June 2020
Colors on shelf tags and colored
feedback stimulating healthy
choices in the grocery store
INVESTIGATING THE MODERATING ROLE OF GENERAL HEALTH INTEREST
ON THE EFFECTIVENESS OF NUTRI-SCORE COLORED SHELF TAGS AND
NUTRI-SCORE COLORED FEEDBACK.
2
Colors on shelf tags and
colored feedback stimulating
healthy choices in the grocery
store
Investigating the moderating role of general health interest on the effectiveness of Nutri-score colored shelf tags and Nutri-score colored
feedback.
LISA KOK
University of Groningen Faculty of Economics and Business
Msc Marketing Master Thesis June 2020 Lisa Kok Herebinnensingel 9-2 9711 GE Groningen (06) 20630051 l.kok.7@student.rug.nl Student number: 3845079
3
MANAGEMENT SUMMARY
Obesity is a rising problem for the public health. Consumers often purchase unhealthy items in the grocery store without knowing how unhealthy the item is, because there is an
information overload. Many consumers do not take the time to read the nutrition facts panel on the back of each package and compare this with other products in the same category to pick the healthiest option. Therefore, two interventions which make it easier for consumers to buy healthy products were tested within this study.
The Nutri-score is an interpretive summary indicator for the healthiness of products. This score indicates to which extent a product is healthy, taking all the nutrition levels into account. The Nutri-score ranges from A to E in which A is healthy and E is unhealthy. The score is combined with colors. For the first intervention, these colors were used during the study to change the colors of shelf tags in the grocery store. The shelf tag of each product had its corresponding Nutri-score color. Furthermore, consumers could easily receive feedback while grocery shopping within this study. This could in practice for example be done via self-scanning devices. For the second intervention, the screen turned the color of the Nutri-score including a sentence whether something was a healthy choice or not after selecting the product. Moreover, both interventions could lead to a boost in effectiveness because
consumers are confronted multiple times with their own behavior. Some consumers are more than others interested in eating healthily. It was expected that these consumers would feel extra guilt when selecting a product that is unhealthy. Therefore, general health interest was expected to positively influence the effect of both interventions on the healthiness of choices.
The study was conducted using a 2x2 between-subjects lab experiment. Participants were recruited via the author’s personal network. There was a total of 643 participants. Participants were randomly allocated to one out of four conditions: Control condition, colored shelf tags, feedback or both interventions. Participants had to select a total of 18 products and had to choose between four products each time. After selecting the products, participants answered some demographical questions.
4 Furthermore, the participants’ general health interest was measured and it was found that this had a significant direct effect on the mean Nutri-score participants chose. However,
participants’ general health interest did not significantly influence the relation between the interventions and the mean Nutri-score. This is probably because these participants always choose healthy, also without the interventions.
5
TABLE OF CONTENT
Management summary ... 3 Introduction ... 6 Theoretical framework ... 11 Food choices ... 11 Nudging ... 12 Nutrition labels ... 13 Shelf tags ... 14 Real-time feedback ... 16Mere-exposure effect and communication channels ... 17
General health interest ... 18
Conceptual model ... 19
Research design ... 21
Research method ... 21
Data collection and participants ... 22
Measures ... 23 Analysis method ... 24 Results ... 27 Descriptive statistics ... 27 Hypothesis testing ... 28 Two-way ANOVA ... 28 Two-way ANCOVA ... 29
Hayes’ Process Macro ... 29
Hayes’ Process Macro with covariates ... 32
Follow-up analysis ... 33
Discussion ... 35
Theoretical contributions and managerial implications... 36
Limitations ... 38
Recommendations for future research ... 39
Conclusion ... 40
References ... 41
Appendices ... 50
Appendix 1: Experiment ... 50
Appendix 2: Health interest scale ... 68
Appendix 3: ANCOVA ... 69
Appendix 4: Process Macro with covariates ... 70
6
INTRODUCTION
Obesity is one of the costliest health diseases of the 21st century (Thorpe, 2009). More than 1.9 billion adults are overweight and 650 million of these people are obese (World health Organization, 2018). The consequences of obesity are a higher risk on cardiovascular
diseases, which were the leading cause of death in 2012, diabetes, musculoskeletal disorders and some cancers (World Health Organization, 2018). However, obesity and overweight can be prevented if consumers change their unhealthy food consumption behavior. Consumers regard being healthy as a fundamental thing to their own well-being (Moorman & Matulich, 1993). Even though consumers have this natural motivation to live healthy, obesity rates are still rising (Wise, 2014). Consumers continuously experience a self-regulation dilemma between instant gratification and the long-term goal to live healthy (Wansink & Chandon, 2006). This may cause dependencies between product choices in the grocery store. It is therefore important to search for ways in which the long-term goal wins from instant gratification.
There are several nutrition labels that try to help consumers making well informed choices about their food consumption like scoring labels, traffic light systems or healthy choice labels. The effectiveness of these labels can be questioned. Some studies suggest that they have a significant effect (Nikolova & Inman, 2015), whereas others suggest that they help consumers identify healthier options, but this does not directly translate in healthier behavior (Ikonen et al., 2019). The effectiveness of these labels has been tested when it was depicted on the front of the packages of products and on shelf tags, but it might also be effective to give the shelf tags a color based on the healthiness of each product. A consumer cannot process all the information he is exposed to in the limited time when making the decision to add or not to add a product to their basket (Hoyer, 1984; Petty, Cacioppo & Schumann, 1983; Gerrior, 2010). According to Bartels et al. (2018) more effectively enabling nutrition information to catch the eyes of shoppers at the point of purchase need further investigation. Specifically with the goal to increase consumer exposure to the information. Enhancing shopper education also needs further investigation according to Bartels et al. (2018).
7 different visual designed label to make sure it stands out. Taking this into consideration, it might be easier and more effective to depict a color on the shelf tag instead of on the package because this could increase the salience due to the bigger display size and the less competing visual environment.
Front-of-package labels are not regulated and there are hundreds of different systems in the marketplace, which are developed by food manufacturers, retailers, and nonprofit
organizations (European Food Information Council, 2018). The existence of many different front-of-package labels creates confusion and limits the effectiveness for consumers to evaluate and compare the nutrition levels of products (Hodgkins et al., 2012). Most labelling systems use varying industry-established nutrition criteria to rate products that carry their front-of-package label rather than objective science-based nutrition criteria which are
consistent with recommendations from the government. This results in front-of-package labels which are applied to products of dubious nutritional quality (Brownell & Koplan, 2011). To make sure that every product has the same type objective nutrition label, it would be
necessary to regulate a certain system via the government. It is more practical for retailers to implement a shelf tag nutrition labelling system compared to front-of-package nutrition labels because there are no changes needed for the lay-out of packages. The feasibility of
implementing front-of-package nutrition label interventions is therefore much lower than shelf tag nutrition labelling.
8 (Sacks et al., 2009). This research will investigate the effectiveness of using the colors of the Nutri-Score as background color on shelf tags including the corresponding letter.
Figure 1. Nutri-Score
Besides the nudges a grocery store can give on beforehand, providing feedback after consumers selected a product is also an option to nudge consumers. Healthy eating
interventions that use behavioral change techniques such as self-monitoring and feedback can strongly influence the degree to which consumers choose healthy (Gustafson & Zeballos, 2019; Helander et al., 2014). A way to give a consumer feedback during their shopping trip is by using the self-scanning device or the website when shopping online. Consumers could see the background of the screen displaying the related Nutri-Score color and letter every time they have selected a product. The exposure to negative feedback might result in a negative self-concept. Consumers have the need to immediately restore their positive self-concept (Kunda, 1990; Dunning & Cohen, 1992), which can be done by selecting a healthy product. Positive feedback gives the intrinsic motivation a boost (Burgers et al., 2015) and could therefore result in selecting more healthy products.
It is expected that these interventions independently cause an increase in sales of healthy products. However, due to the repeated exposure, it is expected that combining the Nutri-score on the shelf tag and the real-time feedback after selecting a product will result in a stronger increase in sales of healthy products. This effect is expected because people are reminded twice that they choose (un)healthy, instead of once. The mere-exposure effect suggests that a consumer develops a positive attitude towards certain stimuli when he is repeatedly exposed to it (Zajonc, 1968).
Nutri-9 Score colored feedback to the background after selecting products on the sales of healthy products?
Consumers tend to be consistent in their behavior (Newcomb, 1953; De Mooij & Hofstede, 2011). Therefore, it is expected that the general health interest of consumers has a positive effect on these relationships. Consumers with a higher level of general health interest are expected to be influenced by the colors more easily. The effect of the colors is therefore expected to be stronger for consumers with a higher level of general health interest. For the combination of the two interventions, it is expected that consumers with a higher level of general health interest will feel extra guilty about their choices, since they are being primed with their bad decisions more often than when they are exposed to it only once. Therefore, a final research question is: How are the effects of colored shelf tags, feedback and the
combined effect on sales of healthy products influenced by consumers’ health interest?
This study will contribute to the existing literature because the effectiveness of the components of an existing label are tested in two new surroundings. Consumers might respond differently to the colors of the label in this context because they are exposed to the color in different parts of the buying process and are used in a different manner. Another important contribution of this study is that it is studied whether the effect is stronger when the interventions are combined, because consumers are than confronted with their behavior more often. The effect might also be stronger for people who have higher general health interest. These people are easier influenced by the colors because they care more about the healthiness of their groceries. The results might be useful for grocery store chains who are looking for ways to promote their customers’ well-being and for policy makers and consumers who are trying to decrease and prevent the obesity problem. Finally, this study extends past research on promoting healthier food choices using nutrition labels, since the literature within this topic is somewhat contradicting.
11
THEORETICAL FRAMEWORK
The healthiness of the choices a consumer makes during their grocery shopping trip is the central concept of this study. This may be influenced by the Nutri-Score on the shelf tag, feedback after selecting a product or the combination of these interventions. The level of health interest might strengthen this effect. These relations will be explained based on reviewed literature.
Food choices
Before making the decision to purchase a certain product, consumers form expectations about the quality of that product (Grunert, 2002). Many quality aspects of a product (such as for example taste) cannot be evaluated by a consumer before purchasing the product. Therefore, consumers rely on other information to form expectations such as the colors of the product, the brand it carries or the packaging of the product (Grunert et al., 1996). The consumer is able to evaluate the quality of the product after the purchase has been made. The gap between the expected and the perceived quality determines the consumer satisfaction about the product and the probability on repeated purchases (Oliver, 1993). When products are not branded, food labels can give consumers the information needed to make an inference about the product (Grunert, 2002).
Consumers are increasingly paying attention to the healthiness of the products they purchase (Nikolova & Inman, 2015), but consumers nowadays have to process much information to decide whether a product is healthy. Most consumers are not able to use all the available information to make a well-informed choice, because a consumer has limited attention and cognitive resources (Houdek et. al., 2018). Besides, most consumers do not have the time to review the nutrition fact panel in detail when shopping for groceries (Block & Peracchio, 2006; Graham et al., 2012; Gerrior, 2010). Furthermore, the healthiness of a product is not a clearly defined term. It is therefore hard for consumers to evaluate whether a product is healthy or not (Brug et al., 1998). Numeric information, such as the number of calories a product contains, is harder to process for a consumer and might backfire if the product contains less calories than expected (Downs et al., 2009). Consumers are often unfamiliar with calories and are unaware of how many calories they should consume daily (Blumental & Volpp, 2010). Symbolic information is based on pictures, colors or symbols instead of
12 compared to numeric information (Ellison et al., 2014). Therefore, this study will be
conducted using symbolic information instead of numeric information.
Nudging
Nudging is a manner to deal with the consumers’ information overload. Nudging is defined as the use of subtle cues in the physical choice context that alters people’s behavior in a
predictable way, without forbidding any options or significantly changing their economic consequences (Thaler & Sustein, 2008). For a nudge to be successful, it should maintain freedom of choice and be transparent to the consumer (Thaler & Sunstein, 2008). A nudge is an intervention which is easily accepted by consumers and not as intrusive for consumers as for example eliminating choice or financial (dis)incentives (Griffihs & West, 2015). Financial incentives have the disadvantage that they could lead to increased caloric intake. The discount a consumer gets leads to a higher total number of total purchased products, which results in an equal proportion of healthy products. The saved money from the cheaper healthy products is spent on more products and therefore more calories (Waterlander et al., 2012).
Within the food domain, there are three categories of nudges distinguished: cognitive oriented nudges, affectively oriented nudges and behaviorally oriented nudges (Cadario & Chandon, 2019). Cognitive oriented nudges are mostly about informing consumers. Affectively oriented nudges focus on the hedonic consequences of eating the food. Behaviorally oriented nudges impact consumer’s behavior without influencing what they know or feel. This research will focus on cognitively oriented nudges. These types of nudges are the most feasible to
implement (Cadario & Chandon, 2019). It is interesting to investigate in which surroundings these nudges are more or less successful, and for which type of consumers because these nudges could then be implemented in a more effective way.
13 et al., 2016). Point-of-purchase food shopping educational interventions are interventions which utilize the food store environment to promote healthful purchasing (Glanz & Yaroch, 2004). There have been numerous studies suggesting that point-of-purchase food shopping interventions have a positive effect on sales of healthy products such as vegetables and fruits (Milliron et al., 2012; Lang et al., 2000; Glanz & Mullis, 1988).
Nutrition labels
A popular type of cognitive nudge is the usage of nutrition labels. There are several studies about food labels and their effectiveness, but these studies are very contradicting. Some support that food labels significantly influence sales of healthy items (Nikolova & Inman, 2015; Cadario & Chandon, 2019; Hersey et al., 2013; Melo et al., 2019), whereas others state that they cannot find significant evidence for this claim (Sacks et al., 2009; Ikonen et al., 2019; Freedman & Connors, 2010; Steenhuis et al., 2004).
Consumers are exposed to many different labels nowadays. These labels aim to increase the number of consumers who readily notice, understand and use the available information to make more nutritious choices for themselves and their family (Food and Drug Administration, 2010). There is a distinction made between reductive nutrient-specific labels, interpretive nutrient-specific labels and interpretive summary indicators. The reductive labels reduce the amount of nutrition information compared to what is provided in the nutrition facts panel without offering any interpretation of this information (Talati et al., 2017). The interpretive labels provide greater evaluation of information in the nutrition facts panel, where the interpretive nutrient-specific labels add an interpretation of the healthfulness of individual nutrients and the interpretive summary indicator labels give a summary of the overall healthfulness of a product (Talati et al., 2017). The latter is the easiest to understand for a consumer (Ikonen et al., 2019; Ducrot et al., 2015). An overview of these labels can be found in table 1 (Ikonen et al., 2019).
Labels Meaning
14 Interpretive nutrient-specific labels Provide greater evaluation of information in the
nutrition facts panel, add an interpretation of the healthfulness of individual nutrients.
Interpretive summary indicators Provide greater evaluation of information in the nutrition facts panel, give a summary of the overall healthfulness of a product.
Table 1: Nutrition labels
An example of an interpretive summary indicator label is the Nutri-Score (Chauliac, 2018). The Nutri-Score has been developed by the French ministry of health and gives consumers an indication of how healthy a certain product is. The Nutri-Score ranges from A to E in which A is the healthiest option and E the unhealthiest option. These letters are combined with traffic light colors. The Nutri-Score label is shown in Figure 1 in the introduction section and in table 1. Research has shown that the Nutri-score appears to be well perceived and understood (Chantal et al., 2017). The label has the greatest effect on perceived healthiness when compared with the nutrition facts panel and the multiple traffic light (Hagmann & Siegrist, 2020). It was therefore chosen to use the letters and colors of this label during the study.
Shelf tags
Besides front-of-package labeling, there are other parts in the customer journey that could result in less consumption of unhealthy products. Many nudges do not only include a sign on a package, but it could for example also include changing the visibility of a product by placing it on eye-level shelfs. It has been proven that for consumers low in price
consciousness, the prominence of unit price information has a positive effect on awareness and usage of such information (Miyazaki, Sprott & Manning, 2000). Since this effect exists for consumers who do not instantly care about the price, it might also be effective to present other information on shelf tags such as nutrition information. Over 50% of study participants of reviewed studies by Hersey et al. (2013) stated that they use or are likely to use shelf nutrition labeling systems and allow these labels to influence their purchase decision.
15 show that attention capture is better with doubled display size. By using a shelf tag to inform consumers about the healthiness of a product instead of the front of a package, salience could increase. This because environment in which the information is then depicted is less crowded and the display size is larger.
The effectiveness of nutritional claims (highlighting whether the food product was low or reduced in fat, cholesterol, sodium, or calories) on shelf tags has been tested before and it was found that this had a significant effect on sales of healthy products (Teisl et al., 2001;
Sütterlin & Siegrist, 2015). A somewhat similar indicator as the Nutri-Score is the American NuVal score (a 1-100 numeric score derived from a nutrition-profiling algorithm, depicted on shelf tags). This score affected high income households and households with children and shifted their yogurt and frozen dinner purchases to more healthful products (Melo et al., 2019). Furthermore, Nikolova & Inman (2015) demonstrated implementing a point of sale nutrition rating system in grocery stores is an effective way in which consumers can ensure that they purchase healthy products. These types of simplified nutrition information systems can help consumers make healthier choices.
Since these forms of point-of-purchase education on the shelf tags are effective, it might also be effective to give the shelf tag a color. Research has shown that the effectiveness of the Nutri-score can be maximized when depicted on every product in store (Hagmann & Siegrist, 2020). Implementing a label in all factories producing product packages has a lower level of feasibility. However, implementing this label on the shelf tags of grocery stores is feasible. The Dutch grocery store Albert Heijn already has a sugar index shelf tag for the soda and dairy sections (Albert Heijn, 2018). This is a separate tag on the shelf which is located next to the price. The label shows clear resemblance with Nutri-score labelling, as it indicates to which extent the product contains sugar combined with a dark red, red, orange or yellow stripe. A similar system could be used for all the other products in the grocery store using the colors of the Nutri-score of each product on the shelf tags. Earlier research has shown that depicting a nutrition label on the shelf tags can have a significant effect (Melo et al., 2019; Teisl et al., 2001), so it is expected that these colors could have a similar influence. Therefore, the first hypothesis is as follows:
16 Real-time feedback
Information can be presented to consumers in many ways. Besides the information a consumer gets before selecting a product, they could also receive some information after selecting a product. Brug et al., (1998) confirm that computer-generated individualized feedback can be effective in changing consumers behavior to better dietary habits and that repeated feedback can increase the longer-term impact of nutrition education on fat reduction. This can be explained because positive feedback primes and stimulates engaging in certain healthy behaviors and is immediately rewarding to consumers (Schwartz, 2018). Positive feedback boosts intrinsic motivation through competence and autonomy needs (Burgers et al., 2015). When consumers are faced with positive feedback while selecting healthy products, this boosts their intrinsic motivation to eat healthily and it could therefore result in adding more healthy items to the basket.
A psychological effect that needs to be considered in this study is the licensing effect. The licensing effect is defined as a prior choice which boosts a positive self-concept licenses the choice of a more self-indulgent option thereafter (Khan & Dhar, 2006). This effect suggests that adding a healthy product to the shopping basket will be followed by a more self-indulgent choice such as an unhealthy product. However, the licensing effect could be minimized using nudges to steer consumers into the right direction with their product choices. Choosing for a healthy product could result in feelings of pride, which leads to an increased likelihood to use self-control to resist temptation (Hofmann & Fisher, 2012). Feelings of pride can be primed via positive feedback. This could therefore also result in selecting more healthy items.
17 and the desire to have a positive self-concept after negative feedback could therefore be a motivator for consumers to purchase healthier products.
An unhealthy product choice might result in feelings of guilt. Feelings of guilt result in an increase in self-regulatory goal importance (Hofman & Fisher, 2012). Feelings of guilt can be primed via negative feedback. This way, consumers are confronted with their unhealthy behavior, causing a threat to their positive self-concept and prevent the consumer from licensing behavior. This results in the following hypothesis:
H2: Nutri-score colored feedback after selecting products from the grocery store increases sales of healthy products.
Mere-exposure effect and communication channels
Both interventions are expected to help consumers make healthier choices in the grocery store when one of them is shown. However, the interventions could both be shown to consumers to boost this effect. The mere-exposure effect suggests that a consumer develops a more positive attitude towards a certain stimulus when he is exposed multiple times to this stimulus (Zajonc, 1968). When both the shelf tag and the feedback show a consumer that he is making an
unhealthy choice, this may result in a worse feeling about adding the product to their basket compared to when they are faced with the fact that they are making an unhealthy choice once. Besides, repeating a claim results in an increased acceptance (Gilbert et al., 1990). Due to the different ways of communicating the same message, the vividness of the message increases for the consumer, making the message more proximate and concrete in a consumer’s mind (Herr et al., 1991). This creates the expectation that the sales of healthy items will increase more when the two interventions are combined, compared to when one of these interventions is used.
The effectiveness of mere exposure has already been proven in several other dimensions. This was for example done in advertising. The extent to which channels evoke responses, differs over channels because of differences in communication power. Therefore, different channels complement each other in the route to persuasion (Dijkstra et al., 2005). The weakness of one channel, for example that consumers may not consciously look at the shelf tags, is
18 occur when channels complement each other. This may also be the case when consumers see the Nutri-score of each product multiple times. This results in the following hypothesis: H3: The effect of Nutri-score colored shelf tags on sales of healthy products is more pronounced when combined with Nutri-score colored feedback.
General health interest
Some consumers are, more than others, interested in their own health. Health interest is defined as the importance of health and taste characteristics of foods in relation to food choice (Roininen et al., 1999). Health-conscious consumers and consumers who have family
members on special diets are more likely to purchase foods indicated as healthy by front-of-package and shelf labeling systems than price-focused consumers (Schucker et al., 1992; Vyth et al., 2010). This could be explained by the elaboration likelihood model (Petty et al., 1983). This model suggests that highly involved consumers take the central route (high elaboration) for processing and low involved consumers take the peripheral route (low elaboration). Which route a consumer takes depends on their degree of involvement. This depends on consumers’ motivation to engage in the desirable behavior and their ability to process the information. For this study, it can be concluded that consumers with a higher general health interest are highly involved compared to consumers with a lower interest and therefore process the information more thoroughly than low involved consumers. High general health interest consumers may give the information more attention than consumers with a lower general health interest.
Besides, consumers with a higher general health interest are more convinced that they should purchase healthy products compared to consumers with a lower level of general health
19 H4: The positive effect of Nutri-score colored shelf tags on sales of healthy products is
stronger for consumers with a higher level of health interest.
H5: The positive effect of Nutri-score colored feedback after selecting products on sales of healthy products is stronger for consumers with a higher level of health interest.
It is proposed that the stronger effect of the combination of colored shelf tags and colored feedback after selecting each product, is positively influenced by the personal health interest of a consumer. The repeated exposure to their bad choices may create extra feelings of guilt compared to consumers with a lower level of general health interest. This is expected because a consumer with a higher health interest level has the motivation to eat healthy. Choosing something unhealthy is in contradiction with their usual beliefs, creating a natural feeling of guilt. If their unhealthy choices are emphasized multiple times, their feelings of guilt might increase even more, resulting in healthier choices. Besides the exposure to negative feedback, the exposure to positive feedback creates a bigger boost to the intrinsic motivation to eat healthy of the consumer with a higher health interest level compared to the consumer with a lower health interest level.
Besides, considering the elaboration likelihood model (Petty et al., 1983), consumers with a higher level of general health interest would take the central route. Therefore, these
consumers process the shelf tag information and feedback more thoroughly than consumers with a lower level of general health interest. When this information is shown twice to these respondents, processing is even deeper than when the information is shown once. Synergies could occur due to the multiple exposure and these are stronger for consumers with a higher level of involvement (Dijkstra et al., 2005). Consumers with a higher level of general health interest are more involved than consumers with a lower level of general health interest. It is therefore expected that these consumers will be influenced more by the combination of the interventions compared to the consumers who score lower on health interest.
H6: The positive effect of the combination of colored shelf tags and colored feedback after selecting each product on sales of healthy products is stronger for consumers with a higher level of health interest.
Conceptual model
20
21
RESEARCH DESIGN
Research method
The experiment was conducted using an online 2x2 between-subjects design. Respondents were randomly allocated to one out of four conditions: 1) shelf tag, 2) feedback, 3) shelf tag + feedback or 4) none (control group). The participants received a situation sketch in which they had to imagine that they were shopping in a grocery store. 4 different products were shown in 18 different product categories. These products were selected based on their Nutri-scores. Only product categories which have varying Nutri-scores were selected for the experiment. At least three of the four products in each question have a different Nutri-score. A complete overview of the experiment can be found in appendix 1. A pre-test was conducted to make sure that participants chose somewhat equally different products This pre-test was done using the control condition with N=10 participants. Product categories in which 70% or more participants chose the same product were deleted or the price of the products in this question was changed in such a way that they were more equal to the other products. This was done because prices were not of importance for this study but could influence decisions to choose one product over another. Prices were based on the prices of a Dutch grocery store and were the same in every condition.
The participants in the shelf tag condition saw colored shelf tags instead of white ones as depicted in figure 3. The colors were based on the corresponding Nutri-score and the letter of the Nutri-score was depicted in the corner of the shelf tag in the shelf tag condition. This letter is not depicted on the shelf tag in the control condition.
Figure 3: Shelf tag condition - Control condition
Participants in the feedback condition saw one colored screen after selecting each product (as depicted in figure 4). The screens correspond with the Nutri-score of each product and include the Nutri-score and color of the Nutri-score. Besides, a small legend is included and a
22
Figure 4: Feedback screens
For the shelf tag + feedback condition, participants saw the colors on the shelf tag and after selecting a product. The participants in the control condition saw no colors on the shelf tags (as depicted on the right in figure 3) and participants did not see the colored screens in figure 4. To make sure that the participants understand what the colors mean in colored shelf tag conditions, a small legend is provided below the choice options explaining the meaning of the colors. This is done in a similar way as in the grocery store (Figure 5).
After selecting the products, participants’ general health interest was measured. For this measure, the existing scale of Roininen et al. (1999) is used. Besides the information needed for the hypotheses, some demographical questions were asked. This was done to analyze how the sample was divided and to take control variables which may affect the outcome into account. Respondents also had to indicate to which extent they were hungry at the moment, so it could be tested whether this had any effect on the choices. Finally, an attention check was included asking the respondents whether they saw the colored shelf tags and feedback.
Data collection and participants
To test the effectiveness of colored shelf tags and feedback, an online experiment was
conducted with 643 respondents. The participants for this study were reached via the authors’
23 personal network and via different social media platforms such as Facebook, WhatsApp and LinkedIn. The experiment was translated in Dutch because most of the population within this personal network are Dutch native. The participants were selected randomly and allocated to one of the four conditions. Participants had incentives to participate in the experiment and could win one out of three gift cards from bol.com worth €30, €20 and €10. The data has been collected between April 22nd and April 26th. The aim was to have at least 50 respondents for each condition to ensure statistical power, which means at least 200 respondents in total. 518 out of the 643 participants finished the experiment. The participants who did not finish the survey were deleted from the dataset. N=518 valid responses remained. The average age of participants was 37 years and ages ranged from 15 to 82 years. 82% of the participants are female and 18% male. 43% of the participants is higher educated (Bachelor’s degree or higher) whereas 57% is lower educated.
Measures
The independent variable in this study is the condition to which the participant is randomly allocated. The dependent variable is the extent to which the participants choose healthy products. This effect may be positively moderated by the general health interest of each participant.
The extent to which participants choose healthy products is measured based on the
corresponding Nutri-score of each product (Chauliac, 2018). The Nutri-score of each product was calculated using the nutrition information on the website of a Dutch grocery store and an excel sheet Nutri-score calculator. An A is healthy (1) and an E is unhealthy (5). It was chosen to use the numbers 1 to 5 to rate the healthiness of each product because the colors of the Nutri-score are used in the conditions. By using the corresponding numbers, it can be measured how the respondents responded to these colors. This could probably not be done by using a measure such as calories, because calories might be higher for a product with an A Nutri-score compared to products with a lower Nutri-score, since the Nutri-score takes multiple nutrient levels into account. This way, it would be harder to measure the
24 The general health interest is measured by asking the respondents to which extent they agree with several nutrition and health related statements. The existing scale of Roininen et al. (1999) is used to measure this. This scale includes statements such as: “I eat what I like and I do not worry much about the healthiness of the food”. The complete scale can be found in appendix 2. The scale was translated in Dutch because the participants of this study are Dutch natives. The last statement (I do not avoid foods, even if they may raise my cholesterol) was not used because this item had the least correlation with the other statements in the original paper of Roininen et al. and it is expected that the term cholesterol is not understood in the same way by every participant. The reversed statements of the scale were recoded for further analysis.
Furthermore, it was measured to what extent participants liked the way in which products were presented and whether they found the number of times they were confronted with the healthiness of products annoying. This was done to test differences within conditions. The degree to which participants were conscious about the level of healthiness of the products they have chosen was also asked. Finally, some covariates were considered. The exact questions of the experiment can be found in appendix 1. The covariates considered were gender, age, education level, allergies, diet, hunger, sport, importance of taste while selecting products and importance of price while selecting products. These covariates were chosen because they could influence the degree to which participants chose healthy or not. For example, participants who are hungry more often purchase products on impulse and these participants could therefore select less healthy products than participants who are not hungry during the experiment (Youn & Faber, 2000).
Analysis method
For analyzing the data, SPSS statistical software is used. The data had to be cleaned before analyzing. Besides, the data must be checked on outliers and missing values. There were 125 respondents deleted from the dataset since these respondents did not finish the experiment. Furthermore, the data was checked on missing values and outliers, but these have not been found since all questions (except the e-mail question to win one of the gift cards) were
25 shelf tags and whether the participant did or did not see feedback after selecting each product. Next, the product choices were recoded into their corresponding Nutri-scores. The mean of these Nutri-scores was calculated for each participant to measure the overall healthiness of their product choices. The allergies of respondents were edited because some of the
participants filled in allergies which were not food related. These answers were changed into no allergy. Besides, some participants filled in allergies for food-related products, but these were products that were not included in the experiment such as fish allergy. These answers were also changed into no allergy. There was a dummy created for dieting participants.
Thereafter, a factor analysis was performed to summarize the participants’ general health interest into less variables. The correlations were analyzed and all correlations gave a significance level of <0.00. The Kaiser-Meyers-Olkin (KMO) statistic equals 0.805 and Bartlett’s Test of Sphericity gave a significance level of <0.00. Since all correlations and Bartlett’s Test of Sphericity were significant and the KMO is >0.5, it can be concluded that a factor analysis is appropriate (Malhotra, 2010). The factor analysis indicated that one or two factors would be the most appropriate. It was chosen to summarize the variables into one factor. This choice was made based on the scree plot of the Eigenvalues, which indicated a nod in the elbow at two, meaning that one factor would be appropriate. The Cronbach’s alpha for one factor equals 0.747, which is >0.6 and can therefore be classified as reliable
(Malhotra, 2010). Besides, the Cronbach’s alpha if item deleted was analyzed and this indicated that the Cronbach’s alpha is at its highest with all current items included.
After this, it can be tested whether there are differences between the four condition groups using a two-way ANOVA test. This test was chosen because the independent variables are categorical variables, the dependent variable is continuous and there was an independence of observations, meaning that participants were allocated to only one condition. The ANOVA indicates whether there are significant differences between groups. Thereafter, the
homogeneity of variances was checked using the Levene statistic which should be >0.05 (Malhotra, 2010). Besides the ANOVA, an ANCOVA was performed to take the control variables into account. After this, the means of both conditions are described.
26 performed. If one of the moderation effects is significant, the effect sizes are determined using model 1 of the process Macro of Hayes. After this, the process Macro of Hayes was
27
RESULTS
The hypotheses were tested using SPSS 26 statistical software. Before analyzing the data, the data was recoded and cleaned as described in the method section.
Descriptive statistics
Before testing the hypotheses, some descriptive statistics were considered. The participants had a mean score of 4.3 on general health interest. Most of the participants sport 2-3 times per week. Taste was an important reason for the participants to choose for a specific product. The mean importance of taste, price and healthiness were measured on a scale from 1 to 7. The mean importance of taste during the product selection equals 6.02. This was the most
important factor. The healthiness was also a reason for respondent to select specific products. The mean importance of healthiness during the product selection equals 5.13. Price was the least important aspect of these three during the product selection. Participants indicated price importance with 4.94 on average. Overall, participants were not very hungry during the experiment. The mean level of hunger was 3.21 on a scale from 0 to 10.
Participants found the number of times they were confronted with the healthiness of products not very annoying. The mean of this measure was 3.06 on a scale from 1 to 7 and did not significantly differ between conditions. They did indicate that they were very conscious of how healthy the products they chose were with a mean of 5.17 on a scale from 1 to 7. This number did significantly differ between groups. The mean health consciousness of product choices was 4.97 for the control condition, 5.05 for the shelf tag condition, 5.29 for the feedback condition and 5.38 for the both condition. Participants evaluated the way in which the products were presented as pleasant and this was rated with a mean of 5.43 on a scale from 1 to 7. This number did not significantly differ between groups.
The mean overall Nutri-score was 2.64. The difference in Nutri-scores per condition are visualized in figure 6. This figure indicates some difference in mean Nutri-scores per condition. It provides some initial support confirming H1, H2 and H3. However, this
28
Figure 6: Mean Nutri-scores per intervention
Hypothesis testing Two-way ANOVA
The first three hypothesis were tested using a two-way ANOVA with interaction. The
independent variables in this model were the shelf tag condition and feedback condition. The dependent variable in this analysis is the mean Nutri-score. The results of the ANOVA are visible in table 2. From these results, it becomes clear that both the shelf tag condition (F(1,514)=6.259; P=0.013) and feedback condition (F(1,514)=38.29; P<0.00) significantly changed the mean Nutri-score of all selected products. The mean Nutri-score within the control condition equals 2.792 (SD=0.362). For both the shelf tag and the feedback condition, this mean was lower. The mean Nutri-score within the shelf tag condition equals 2.679
(SD=0.409). For the feedback condition, the mean Nutri-score equals 2.557 (SD=330). This is in line with hypothesis 1 and 2. The Levene’s test indicated a significance level of P=0.145, meaning that the assumption of homogeneity between groups is not violated. From this, it can be concluded that both hypothesis 1 and 2 are confirmed.
Source Type III Sum
of Squares
df Mean
Square
F Sig. Partial Eta
Squared Corrected Model 6.255 3 2.085 15.079 .000 .081 Intercept 3584.829 1 3584.829 25928.139 .000 .981 Shelf condition .865 1 .865 6.259 .013 .012 Feedback condition 5.295 1 5.295 38.297 .000 .069 2,3500 2,4000 2,4500 2,5000 2,5500 2,6000 2,6500 2,7000 2,7500 2,8000 2,8500
No shelf tag Shelf tag
29 Shelf condition * Feedback condition .133 1 .133 .960 .328 .002 Error 71.066 514 .138 Total 3675.744 518 Corrected Total 77.320 517
Table 2: Two-way ANOVA
The interaction effect indicates a significance level of (F(1,514)=0.960; P=0.328), which is not significant. However, when looking at the mean Nutri-score of the participants in this condition, it becomes clear that participants chose the healthiest in this condition with a mean Nutri-score of 2.507 (SD=0.384). From this, it can be concluded that the conditions do have a summed effect, but the effect is not moderated when the conditions are combined. Therefore, it can be concluded that there is not enough evidence to confirm H3.
Two-way ANCOVA
Besides the ANOVA, an ANCOVA was performed to control for effects of other variables. The variables that functioned as covariates in this analysis were gender, age, sport, education level, diet, allergies, hunger, importance of taste and importance of price. Inclusion of these covariates, resulted in a higher significance level for the shelf tag condition (F(1,505=6.020; P=0.014) and feedback condition (F(1,505=38.436; P<0.00), but not in a higher significance level for the combination of conditions (F(1,505=0.734; P=0.392). The covariates which had a significant effect (P<0.00) on the mean Nutri-score were gender (β=-0.184), age (β=-0.003), education level (β=-0.041) and sport (β=-0.061). An overview of this analysis can be found in appendix 3. The conclusion from the ANOVA remains the same when controlling for the covariates. As in the previous analysis, H1 and H2 are confirmed, whereas H3 is rejected.
Hayes’ Process Macro
The final three hypothesis were tested using mode 3 of the Hayes’ Process Macro. The direct effect of colored shelf tags on healthiness of choices was already confirmed, but the
participant’s general health interest may positively moderate this effect. First, the VIF statistics were taken into account to make sure that there is no multicollinearity. All VIF statistics were found to be <10, indicating that multicollinearity is not an issue (Malhotra, 2010). All variables were mean centered to perform the analysis. The model summary
30 Nutri-score is statistically significant (t(510)=-12.603; P<0.00). This indicates that the
participants’ general health interest had a significant direct effect on the healthiness of their product choices. Participants with 1 standard deviation higher general health interest, chose approximately 0.186 Nutri-score lower (SE=0.0147). The model is summarized in table 3.
Variable Coefficient se t p LLCI ULCI
Constant 2.6347 0.0144 182.6392 0.000 2.611 2.6585 Shelf tag -0.0976 0.0289 -3.3823 0.0008 -0.1451 -0.05 GHI -0.1858 0.0147 -12.6034 0.000 -0.2101 -0.1615 Shelf * GHI -0.0543 0.0296 -1.8332 0.0674 -0.1031 -0.0055 Feedback -0.1767 0.0289 -6.1227 0.000 -0.2242 -0.1291 Shelf * Feedback 0.0061 0.0577 0.1061 0.9155 -0.089 0.1012 Feedback * GHI 0.0074 0.0295 0.251 0.8019 -0.0412 0.056 Shelf * Feedback * GHI 0.0213 0.0593 0.3595 0.7194 -0.0764 0.119
Table 3: Process Macro
The first hypothesis that is considered is the moderating effect of general health interest on the relation between colored shelf tags and the mean Nutri-score (H4). The interaction of these variables gives a moderately significant effect (t(510)=-1.8332; P=0.067). The corresponding beta gives a coefficient of -0.0543 (SE=0.0296). This indicates that participants in the shelf tag condition with a higher level of general health interest chose a lower mean Nutri-score than participants with a lower level of general health interest, and that this effect is stronger within the shelf tag condition compared to the control condition. From this, it can be
concluded that there is marginally statistical evidence confirming H4.
Since the effect of general health interest on the relationship between the shelf tag
31 between the shelf tag condition and mean Nutri-score was negative but not significant
(β=-0.044 ; SE=0.041; t(510)=-1.076; P=0.28). At the mean general health interest, the relationship was also negative but here it is statistically significant (β=-0.098; SE=0.029; t(510)=-3.382; P< 0.00). Finally, at +1 standard deviation on general health interest, the relationship was negative and significant (β=-0.151; SE=0.041; t(510)=-3.671; P<0.00).
Figure 7: Interaction effect shelf tag condition * General Health Interest
The second hypothesis which is tested within the process Macro of Hayes is the effect of general health interest on the relation between feedback and the mean Nutri-score (H5). This interaction does not result in a significant effect (t(510)=0.251; P=0.8019). The analysis gives a coefficient of 0.0074 (SE=0.0295), but this cannot be interpreted since the analysis is not significant. This indicates that there is not enough statistical evidence confirming H5.
Finally, the hypothesis of the interaction effect of the general health interest on the relation between both the feedback and shelf tag intervention and the mean Nutri-score (H6) is tested. The interaction effect did not indicate a statistically significant effect (t(510)=0.3595;
P=0.7194). The analysis indicates a coefficient of 0.0213 (SE=0.0593), which cannot be interpreted since the analysis is not significant. This indicates that there is not enough statistical evidence to confirm H6.
2,30 2,35 2,40 2,45 2,50 2,55 2,60 2,65 2,70 2,75 2,80 2,85 2,90 0 0 0 M ea n Nu tr i-sc o re
General health interest
32 Hayes’ Process Macro with covariates
The interaction effects of H4, H5 and H6 are tested without controlling for other variables which might influence participants’ choices. Therefore, the previous analysis was also performed while taking the control variables into account. The control variables which were considered are the same as in the ANCOVA, namely gender, age, education, allergies, diet, hunger, sport, taste and price. A complete overview of these results can be found in appendix 4. Multicollinearity was taken into account to make sure that the variables do not overlap. All VIF statistics were <10, indicating that multicollinearity is not an issue. All variables were mean centered while performing the analysis. The overall model is statistically significant (F(16,501=15.450; P<0.00) and explains 33% of the total variance. The only covariate that had a significant effect on the outcome of the analysis was gender (β=-0.1481; P<0.00).
Within the previous analysis, marginally significant evidence was found confirming the moderating effect of general health interest on the relation between the shelf tag intervention and the mean Nutri-score. Adding the mentioned covariates to the analysis makes this effect more significant (t(501)=-1.887; P=0.0597). The coefficient of this analysis equals -.0558 (SE=0.0296). Even though the analysis is nearly significant, the P-value is not below 0.05. Therefore, it can be concluded that there is marginal evidence to confirm H4. However, further research is necessary to draw conclusions about the effect of general health interest on the relation between colored shelf tags and the mean Nutri-score.
H5 was also tested taking the covariates into account. For this hypothesis, it was analyzed whether there is a moderating role of general health interest within the relation between the feedback intervention and the mean Nutri-score. This did not result in a significant effect (t(501)=0.0375; P=0.970). The coefficient of this analysis equals 0.0119 (SE=0.0573). However, this coefficient cannot be interpreted since the analysis is not statistically
significant. Neither the analysis with covariates nor without covariates indicated a significant effect. Therefore, it can be concluded that there is not enough evidence confirming H5, meaning that H5 is rejected.
Finally, the interaction effect of the general health interest on the relation between both the feedback and shelf tag intervention and the mean Nutri-score (H6) has been analyzed. The analysis without covariates did not result in a significant effect. Analyzing the same
33 The coefficient of this analysis equals 0.0316 (SE=0.059), but this cannot be interpreted since the effect is not statistically significant. There is not enough evidence to confirm the
hypothesis. Therefore, H6 has been rejected.
Follow-up analysis
All participants had to indicate whether they saw colored price tags or white ones, and whether they have seen colored feedback screens or not. This was done to check whether participants noticed the manipulations. Some of the participants did not answer this question correctly. They may have not seen the manipulation and therefore respond differently to it. This could result in different results. Therefore, the analyses were also performed while filtering out the participants who did not pass the attention check (N=402). The results of this analysis can be found in appendix 5.
The first analysis to perform is the ANOVA. This analysis indicates a similar significance level for the feedback condition (F(1,398)=29.258; P<0.00), which is in line with hypothesis 2 and the previous performed analyses. However, for the shelf tag condition the significance level is different (F(1,398=3.014; P=0.083), indicating only a marginally significant effect. This is not in line with the previous analysis and indicates that participants may be influenced by the interventions without consciously noticing them or that the effect shrunk due to the smaller sample size. The interaction of both conditions did still not result in a significant effect (F(1,398)=0.975; P=0.324).
34 The Process Macro indicates a significant model summary (F(7,394=26,24; P<0.00). The model explains 31.8% of the variance. The direct effect of the shelf tag condition was found significant within this analysis (t(394)=-2.99; P<0.00) and indicates a significant interaction effect of shelf tag and general health interest (t(394)=2.27; P=0.0235) with a coefficient of -.0791 (SE=0.0348). This is in line with hypothesis 4, but the previous analysis indicated a lower level of significance. This may indicate that the participants whom actively processed the colored shelf tag were more influenced by the intervention compared to participants who did not pass the attention check. The interaction between the feedback condition and general health interest is not significant (β=0.029; SE=0.0337; t(394)=0.8606; P=0.390), which is similar to previous analysis but not in line with H5. Finally, the effect of the interaction including both conditions and general health interest was not significant (β=0.0593; SE=0.07; t(394)=0.8482; P=0.3969) as in the previous analysis, which is contradicting to H6.
Finally, the process macro including covariates was performed. The overall model was statistically significant (F(16,385)=13.0663; P<0.00) and explains 35,2% of the variance. There was one significant covariate within this analysis, namely gender (β=-0.119; SE=0.044; P<0.00). The interaction effect of the shelf tag condition and general health interest was significant (β=-0.0759; SE=0.0349; t(385)=-2.178; P=0.0299) which is confirming H4 and is contradicting to the analysis in which participants’ attention was not considered. The
interaction effect of the feedback condition and general health interest was not significant (β=0.0227; SE=0.0335; t(385)=0.677; P=0.4988) and this is contradicting to H5, but similar to previous analysis. The interaction between both conditions and general health interest (H6) was similar to previous analysis and had no significant effect (β=0.0786; SE=0.0697;
35
DISCUSSION
During this study, the effectiveness of adding the Nutri-score colors to shelf tags and giving consumers Nutri-score colored feedback on their product choices was investigated in relation with the overall healthiness of product choices. Furthermore, it was studied how general health interest influenced this relationship. The first research question was about the effect of adding the Nutri-score colors to shelf tags on the sales of healthy products. From this study, it became clear that participants who saw colored shelf tags chose healthier than participants who saw white shelf tags. The Nutri-score of participants with colored shelf tags was
approximately 0.114 Nutri-score lower. This is in line with hypothesis 1. The second research question investigated the effect of colored feedback after selecting a product. This
effectiveness has also been confirmed. Participants who saw colored feedback after selecting each product chose healthier than participants who did not see feedback. These participants chose a mean Nutri-score that was 0.234 lower compared to participants who did not see the feedback. This is in line with hypothesis 2. The significance of both of these interventions is a contribution to existing literature and could be a step in the battle against overweight.
Thereafter, the third research question about the interaction of both interventions was tested, but this was not found to be significant. In conclusion, both interventions do have affect the healthiness of product choices, but combining the interventions does not boost this effect. This is not in line with hypothesis 3. However, it does not buffer the effect. Implementing both interventions could therefore still be very effective in nudging consumers to make healthier choices. A reason that the effect was not boosted could be that the mere exposure effect did not occur. The mere exposure effect is stronger when there is a delay between exposure and attitude measure (Bornstein, 1989). Within this study, there was no delay between the exposure and attitude measure, which could have resulted in lower effectiveness of the mere exposure. Furthermore, Bornstein (1989) suggests that the mere exposure effect levels off after 10 to 20 exposures. Within the current study, participants had a total of 36 exposures (18 products and exposed to the information before and after selecting a product), meaning that the effectiveness of the mere exposure has probably leveled off after the 10th product choice.
36 interaction effect found. However, for the shelf tag condition, the analysis indicated a
marginally significant effect, indicating that consumers with a higher level of general heath interest were influenced more by the intervention compared to consumers with a lower level of general health interest. For the feedback condition and the condition in which both
interventions were applied, there has no significant effect been found. This indicates that the effectiveness of the interventions is not higher when participants have a higher level of general health interest. A reason for this could be that these participants always choose healthier than participants with a lower level of general health interest, no matter whether there is a nudge to steer them or not. The results of the study are summarized in table 4.
Hypothesis Accepted Rejected
H1: Nutri-score colored shelf tags in the grocery store increase the sales of healthy products.
H2: Nutri-score colored feedback after selecting products from the grocery store increases sales of healthy products.
H3: The effect of Nutri-score colored shelf tags on sales of healthy products is more pronounced when combined with Nutri-score colored feedback.
H4: The positive effect of Nutri-score colored shelf tags on sales of healthy products is stronger for consumers with a higher level of
health interest. *Marginally
significant
H5: The positive effect of Nutri-score colored feedback after selecting products on sales of healthy products is stronger for consumers with a higher level of health interest.
H6: The positive effect of the combination of colored shelf tags and colored feedback after selecting each product on sales of healthy products is stronger for consumers with a higher level of health interest.
Table 4: Conclusions hypotheses
Theoretical contributions and managerial implications
37 be a great step in the battle against overweight. However, the long-term effectiveness of the intervention needs further research.
Second, the study confirms the effectiveness of real-time feedback. The current study investigated the effectiveness after selecting each product, whereas the study of Brug et al., (1998) investigated the effectiveness of feedback about the overall purchases. This is a new finding within literature and could be interesting for grocery stores and policy makers. Furthermore, the study contributes to existing literature due to testing the effectiveness of the combination of the shelf tag intervention and the feedback intervention. Unfortunately, this did not have any additional effect. However, it could still be interesting to implement both within grocery stores. Even though there is no moderated effectiveness of combining the two interventions, they do have a summed effect that increases the healthiness of product choices.
Another new contribution to the literature is the moderating role of general health interest on the relation between the shelf tag and feedback interventions and the mean chosen Nutri-score of participants. This has, to the best of the authors’ knowledge, not been examined before. There appears to be some effect of general health interest for the colored shelf tags, however this effect was only marginally significant. It is therefore interesting to investigate the moderating effect of general health interest with a larger sample size to ensure statistical power. There was no effect found for the moderation of general health interest within the feedback intervention and the combination of the interventions. This study could be replicated with a larger sample size to ensure that there is no effect.
The literature on the licensing effect is extended by the current study (Khan & Dhar, 2006). However no specific tests have been performed with the healthiness of products in the early and later stadium of the shopping trip, it was clear that participants within the conditions with shelf tag or feedback interventions chose a healthier mean Nutri-score. This indicates that the licensing effect has leveled off or maybe did not appear at all as a result of the interventions. Further research is needed to confirm this statement and to get a clear insight in how this effect develops over the shopping trip while such interventions are implemented.
38 experimenting with the intervention of implementing Nutri-scores on their shelf tags and their website (Albert Heijn, 2020). This indicates that grocery stores are interested in marketing of healthier product choices and in their consumers’ well-being. Furthermore, policy makers could implement similar systems to stimulate healthier food choices among a bigger public. If the system is implemented, it could contribute to a higher level of public health and make a difference in the battle against overweight and obesity.
Limitations
Within the study, a few limitations occurred. First, the study was done in an online
environment. Participants may behave differently compared to when this was done in a field study (Aronson et al., 1998). Furthermore, 83% of participants were female. Even though this is not a realistic reflection of the population, most of the females do the main grocery
shopping within a family (van Vliet, 2019). Therefore, this limitation may have affected the external validity to a lesser extent. Furthermore, the sample which was used for the analysis had a size of 518 respondents. This sample size was enough to ensure statistical power and to ensure external validity. However, a bigger sample may be necessary for a better result among the analyses including moderators and to increase external validity. Finally, the internal validity was guaranteed due to the usage of for example an existing scale for measuring general health interest. All questions needed for the analysis were mandatory, causing no missing values.
The prices of products may also have influenced choices in several ways. First, participants had no spending budget. This was done because participants then had to think with their usual budget. However, participants do not really buy the products. This may therefore influence their decisions and make them choose for a product they would normally perceive as too expensive. There was somewhat controlled for this limitation because prices were the same in all conditions.
Furthermore, some of the participants indicated they never buy multiple items within the product options. These participants may have chosen differently than respondents who do buy those products regularly. They may for example have chosen based on price, brand or
39 The way in which products were presented may have decreased the reliability. Products could be presented in a similar way in an online environment, but not in the brick and mortar
grocery store. For the feedback intervention, it might also be hard to represent the information in a similar way. There was an attention check included in the analysis, but 125 participants did not manage to pass the attention check. Participants who were either in the shelf tag condition or in the feedback condition often filled in they had seen both interventions, while they saw only one intervention. This indicates that they did consciously saw an intervention, but did not pass the attention check. The attention check was asked to participants after they have filled in the general health interest scale. Due to this, there was a delay between the last exposure to the interventions and the attention check, which may have resulted in a lower number of participants who correctly answered the question. If the study is replicated in the future, it is recommended to switch the order of these questions.
Recommendations for future research
As mentioned within the limitations section, the study was performed while doing an online lab experiment. The study could be replicated while doing a field study to empower statistical evidence and to make sure consumers also behave this way in real life. Furthermore, the study could be replicated using additional moderators, as general health interest did not appear to be effective for every condition. The effectiveness of general health interest could also be tested while using a larger sample size to make sure that there is an interaction of general health interest on the relation between the shelf tag intervention and the mean Nutri-score and to make sure that there is no interaction for the other conditions. Additional moderators could for example be participants’ education level, social economic status, or customers’ value
orientation (whether they purchase their groceries based on prices, taste or healthiness). The study could also be replicated in another country or in several countries at the same time because there could be cultural differences in responses to the interventions.
40 interaction between the shelf tag condition and the general health interest did become
significant when filtering out these participants. It may therefore be interesting to study the difference in effectiveness of the interventions while consumers consciously processed the intervention versus when consumers did not consciously process the interventions. Finally, it may be interesting to investigate the effectiveness of these types of labels on the long term because the effectiveness could level off when consumers get used to the interventions.
Conclusion
The goal of this research was to find out to what degree point-of-sale health information influences the overall healthiness of groceries consumers choose. From this research, it can be concluded that there are multiple ways to help consumers choose healthier. Both the Nutri-score colored shelf tags and the Nutri-Nutri-score colored real-time feedback made consumers choose a lower mean Nutri-score than participants who did not see any of these interventions. This effect has also been found while controlling for the participants’ gender, age, education level, allergies, diets, hunger, sport and the degree to which they found price and taste important while selecting products. Furthermore, the combination of both interventions resulted in an even lower mean Nutri-score, but this combination did not boost the effect. The effectiveness of the interventions has been proven and found significant, which could have great implications for policy makers and grocery stores, which could in turn have an even more important impact on public health.
Furthermore, a persons’ general health interest seems to positively influence their mean Nutri-scores and the effectiveness of the shelf tag intervention, however further research is needed to confirm this statement, since it was marginally significant. No evidence has been found to confirm that the general health interest positively influences the effectiveness of the feedback intervention or the combination of the interventions. Looking forward, it will be important to find out for which other types of consumers, besides consumers with a higher level of general health interest, are influenced more and less by these types of interventions. This will
