Facing the
Obesity Battle
Intervening Psychological Choice
Mechanisms to Precipitate Health-Conscious
Shopping Decisions
Master Thesis
Facing the Obesity Battle
Intervening Psychological Choice
Mechanisms to Precipitate Health-Conscious
Shopping Decisions
Study in an Online Supermarket
-Author
1
stSupervisor
2
ndSupervisor
Examiner
J. Assenmacher
(S3346846)
prof. dr. ir. K.
van Ittersum
M. T. van der
Heide
dr. J. van Doorn
Jana.Assenmacher@web.de K.van.ittersum@rug.nl M.t.van.der.heide@rug.nl J.van.doorn@rug.nl
Harbigstraße 27 Nettelbosje 2 Nettelbosje 2 Nettelbosje 2
52249 Eschweiler 9747 AE Groningen 9747 AE Groningen 9747 AE Groningen
Germany The Netherlands The Netherlands The Netherlands
+49(0)1731581413 +31(0)50 36 36639 +31(0)50 36 37074 +31(0)50 36 33657
Completion Date
02.07.2018
University of Groningen
Faculty of Economics and Business
II
Preface
Obesity is the second leading preventive cause of death, killing 300,000 Americans a year (Flegal, Williamson, Pamuk, & Rosenberg, 2004). This estimation dramatically underlines the tremendous impact corpulence can have on the lives of individuals all around the globe. Certainly, dying from obesity is an extreme illustration, however, obesity has deep-ranging consequences for the physical and psychological well-being of individuals and can therefore breed a tremendous decrease in lifetime and life quality. Isn’t that shocking? Due to self-made over-consumption decisions, society grows into its own misery. Obesity is clearly a man-made problem (van Ittersum, 2017).
While I definitely enjoyed the cultural experience of spending a year in the US, the health aftermath of my journey was not equally pleasant. Coming back, I had gained a vast amount of weight, just within one year. It was personally shocking that, within such a short amount of time, such weight gain could occur. Ever since, it has been my personal interest to follow a healthy diet and a health-conscious lifestyle. Studying marketing and being taught how to ‘sell products to consumers at all costs’ triggered the thought whether companies’ share in the obesity epidemic is not just as tremendous as our own consumption decisions. Subsequently, it was my personal desire, before commencing my professional career, to acquire knowledge on how to responsibly market products without negative long-term consequences for consumers, which finally steered me to the creation of this thesis.
Most importantly, I wish to thank several individuals who supported and encouraged me during the process of submitting this thesis. First, I would like to express deep gratitude to my supervisors Koert van Ittersum, Martine van der Heide and their colleague Plamen Dragiyski who, not only through their qualitative and helpful feedback, but also moral aid, supported and encouraged me. Furthermore, I owe gratitude to my friends and family who, not only listened to my thoughts and gave me inspiration, but also encouraged me to not lose track of my objective. Lastly, a special thank you to those that revised my thesis and gave critical and truly helpful feedback. Eventually, no person succeeds without an adequate team of coaches in the back. Therefore, my honest gratitude to all those who coached, supported and motivated me and thereby, aided accomplishing this thesis and master. Thank you. Groningen, July 2nd of 2018
III
Management Summary
Individuals’ shopping decisions and psychological choice-making processes inherit one of the biggest health issues of our society- obesity. To conquer this issue, two different concepts of consumer choice-making are taken under consideration in this study. First, the idea of licensing, describing that consumers usually succeed healthy with unhealthy grocery choices, is dissected. Secondly, the what-the-hell-effect is considered, stating that once consumers have abolished their goal of healthy purchases, they persist with doing so. These decision-making phenomena, which prime unhealthy shopping and subsequently consumption-decisions, might induce overweight. The interdependence of shopping decisions that is encompassed in the two psychological conceptions, is well-known and vastly studied, however, a research shortcoming existed regarding the deterioration of the two concepts. Therefore, this paper aimed at discovering possibilities to break the licensing and what-the-hell-effect in food choices and ultimately guide consumers to healthier shopping decisions.
In this regard, it was aimed to vanquish the licensing effect with a concept called goal-commitment, which depicts that an activated health goal steers further healthy decision making and could therefore conquer licensing. Furthermore, regarding the what-the-hell-effect, it was intended to establish how the adverse licensing what-the-hell-effect, following unhealthy choices with healthy choices, can be activated to lead to a cycle of healthy choice-making.
In order to disrupt these psychological concepts and lead to their more advantageous counterparts, two different interventions were installed in an online supermarket experiment with 353 American participants. It was opted for an online store due to the novelty of online grocery shopping and to assess the existence of a differential effect between online and well-established offline interventions. Consequently, it was investigated if a health-directed (highlighting the healthiness of a yogurt) or a taste-stressing (emphasizing the tastiness of a yoghurt) intervention alternates the choice-making process.
IV
evolve as significant. Additionally, the what-the-hell-effect was resilient against the interventions and could therefore not be destroyed. Lastly, it was established that, although it was temporarily possible to break the licensing effect, this influence did not persist until the end of the shopping trip.
On basis of this analysis, certain implications for practitioners can be formulated. Firstly, already humble interventions, like a product label, can guide consumers to healthier decision-making. Companies can facilitate the healthy choice-making process and thereby chaperon the well-being of their customers, which allows them to be labeled as a responsible and sustainable establishment. However, as customer move quickly on the website and the online environment lacks sensorial experiences, online nutrition interventions might differ from their offline counterparts and therefore, need to be evaluated with caution.
Furthermore, health interventions are more effective when connected to healthy previous choices. Thus, as online environments facilitate individual choice interventions, companies ought to keep track of the healthiness of consumer choices and individually target customers of healthy previous choices with a health label. Although the health intervention showed a short-term effect not only on the choice to which is was connected but even stronger on the choice which followed, it lacked long-term consequences. Furthermore, customers in the online supermarket were very price-sensitive. In order to collectively conquer these challenges, managers are advised to apply a loyalty health point system, next to the use of healthy labels. Thereby, customers could collect points and be rewarded for making healthy decisions. Through such a solution, companies can profit through the gathering of customer data and an increasing customer loyalty.
Moreover, managers could profit by continuously updating their customers during the shopping trip about the healthiness of their shopping basket. They could do so by indicating the healthiness through green (when healthy) or red (when unhealthy) in-store icons that signal the customer a healthy or unhealthy current shopping basket. Lastly, through messages directed at customer’s health interest, clients will not only be attracted to the ads and products but will also be reminded to make health-conscious shopping decisions for their personal benefit.
V
Table of Content
Preface ... II Management Summary ... III
Introduction ... 1
Theoretical Foundation ... 3
Consumer Choice-Making ... 3
Supermarket Interventions ... 7
Conceptual Model & Hypothesis ... 10
Hypotheses... 10 Conceptual Model... 12 Methodology ... 13 Research Design ... 13 Data Collection ... 14 Results ... 17 General Analysis... 17 Results Stage 1... 22 Results Stage 2... 28 Results Stage 3... 33 Summary of Results... 35 Discussion ... 36 General Discussion ... 36 Contributions ... 40
Limitations & Future Research ... 41
Implications ... 42
Conclusion ... 45
References ... 46
Table of Appendices ... 54
1
Introduction
Obesity - Rarely a problem in society is simultaneously as common and as severe as the increasing rate of an obese population (van Ittersum, 2017). In fact, between 1980 and 2013, the number of adults worldwide with a Body-Mass Index (BMI) above the critical level of 25 increased by 8% (Ng et al., 2014)1. Not only has obesity deep-ranging consequences for the physical health of the obese, but it also affects the psychological well-being. Obesity causes diabetes and heart diseases through high blood sugar and pressure, which generate an escalation of heart attacks and can cause deadly forms of cancer (Wilding, 2001; Mokdad et al., 1999). Furthermore, victims of obesity face social exclusion and a reduction in their personal self-esteem. Discrimination in various life areas upgrades physiological problems to psychological diseases which, in turn, leads to a tremendous decrease in happiness and life quality (Wardle & Cooke, 2005; Kolotkin, Meter, & Williams, 2001). However, the consequences of obesity do not only affect the individual but also the economy as a whole. Namely, obesity induces an increase in health costs and a decrease in workforce productivity (van Ittersum, 2017). Therefore, obesity is not only a personal challenge but also a widespread economic phenomenon. Urgent global action is needed to conquer this challenge (Finkelstein, Ruhm, & Kosa, 2005).
When inspecting our society, a discrepancy exists between healthy shopping intentions and final – unhealthy – shopping baskets, ultimately leading to obesity (van Ittersum, 2015). One of the causes for this discrepancy is licensing, a concept known in psychology as moral licensing, which has been widely studied in the consumer psychology domain (e.g.Hui, Bradlow, & Fader, 2009; Khan & Dhar, 2006). Licensing in consumption choices implies that, after having purchased a rather healthy item, this licenses or allows the consumer to make an unhealthy subsequent choice. Consequently, consumers’ choices are not independent from each other, but past choice histories affect future decisions (Khan & Dhar, 2006). Moreover, a divergent theory, the what-the-hell-effect, stresses that after buying unhealthy, consumers cancel their goals and keep making unfortunate choices (Polivy, Herman, & Deo, 2010).
Clearly, the issues of obesity as well as the two psychological effects need to be tackled. A potential game changer could evolve through the rise of online retailing, as
2
shopping environments are known to shape shopping behaviors (Odoms-Young, Singleton, Springfield, McNabb, & Thompson, 2016). Online retailing was primarily used for ordering clothes, technical supply or gifts. In recent years, however, the trend of ordering groceries online has expanded (Statista, 2017). Furthermore, marketers have come to the understanding that ‘selling products at all costs’ and neglecting the effect this has on the customers’ lives does not necessarily ensure long-term profitability. Companies are becoming more motivated to create a sustainable and responsible brand image (van Ittersum, 2015). Therefore, it is of high interest, not only for consumers but also for businesses, to examine if there is a possibility to intervene the licensing and the what-the-hell-effect to ultimately guide consumers to healthier choices and possibly decrease obesity. On these grounds, it is intended to discover possibilities to break the licensing and the what-the-hell-effect. For this purpose, it is analyzed how previously made choices influence the reception of the customer for certain interventions and how these interventions influence further decision-making. Thereby, it is aimed to test the following research question:
“Is it possible to break the licensing and what-the-hell-effect in food choices and ultimately lead consumers to consistently make healthier shopping decisions?”
3
The remainder of this paper is organized as follows. In section one, an overview of underlying theories and existing studies is given, which leads to the formulation of the hypotheses and conceptual model in section two. In section three, a detailed description of the research design is specified to elaborate on the specific study undertaken. Subsequently, the results of the study are scrutinized and then discussed in order to finally arrive at theoretical and managerial implications as well as limitations, which might be of interest for further research.
Theoretical Foundation Consumer Choice-Making
4
categories. Through these establishment, crucial proof of the correlation of choices is embodied.
From the Licensing Effect to Goal Commitment. The abovementioned literature commonly refer to the concept of licensing, which was originally explored in the moral domain as moral licensing. There it ‘allows’ people to indulge in a rather immoral task after having fulfilled a good deed (Mazar & Zhong, 2010). However, it has recently also been considered in the food consumption domain (e.g. Khan & Dhar, 2006; Weibel, Messner, & Bruegger, 2014; Ishibashi, Miyazaki, & Katsutoshi, 2015). Licensing, in health matters, suggests that first choosing a virtue choice, promotes the consumer to make a subsequent vice choice (Hui et al., 2009). Thus, when at the supermarket, customers add groceries to their basket, depending on their previous choices (van der Heide et al., 2016). Overall, this process occurs unconsciously, without the active consideration of the consumer (Khan & Dhar, 2006). Indulging in a healthy choice boosts the self-image - the image that a particular person carries of him- or herself- and this boost operates as a buffer against negative emotions of an unhealthy purchase (Khan & Dhar, 2006). Consequently, the consumer is able to decide for an impulsive purchase without experiencing a bad conscious (Kivetz & Zheng, 2006). On these grounds, the licensing effect implies that a healthy choice is usually followed by an unhealthy choice and thus, the healthiness of later choices is reduced (Hui et al., 2009).
Additionally, the theory of goal commitment clarifies that consumers do not show such alternation of choices but rather stay consistent in their behavior (Dhar & Simonson, 1999). Especially after positive experiences, consumers are found to stick to decisions that correspond with their overall goals. Contrary to the licensing effect, this would imply that consumers keep buying healthy, if this corresponds to their overarching objective of a healthy consumption-style (Novemsky & Dhar, 2005). Research describes that goals can be activated unconsciously and that these subconscious primes lead conscious decision-making (Dhar & Simonson, 1999). Once activated, the target promotes behavior into the direction of the overall objective and keeps the individuals from deviation (Novemsky & Dhar, 2005). Even through obstacles and in the presence of attractive choices, consumers follow their objective until accomplishment (Bargh, Gollwitzer, Lee-Chai, Barndolla, & Troetschel , 2001).
5
explicit matter. However, borrowed literature can be used as an adequate baseline for the development of an effective intervention. In order to break well-established habits, which the licensing effect could be considered as, Papies, Potjes Keesman, Schwinghammer and van Konigsbruggen (2014) find that health-priming consumers with diet information, reduces snack purchases among the overweight. Furthermore, customers are not required to be consciously aware of the prime while shopping in order for the effect to occur. Although this study does not directly concern licensing, it shows a way to lead the consumer to healthier choices, congruent with breaking the licensing effect. Furthermore, Verplanken and Wood (2006) propose that for purchasing habits that lead to obesity, “upstream interventions” (Verplanken & Wood, 2006) should be utilized, which should occur before habit performance. By doing so, environmental cues are disrupted and new ones, like the urge to buy healthier, can be established. It is stressed that providing information is insufficient and deeper interventions are necessary, directly at the point of sale. Therefore, intense interventions, which go beyond information provision, are necessary. Lastly, van Ittersum (2017) states that the buying behavior, which contributes to the licensing effect and ultimately to obesity, requires a deep reaching environment that supports behavioral change.
From The What-the-hell-effect to Adverse Licensing. Nevertheless, consumers’ first buying decision can also demonstrate unhealthiness. In line with this remark, another concept, the what-the-hell-effect, was established and demonstrates differential claims (e.g. Polivy et al., 2010; Herman, Polivy & Esses, 1987). Respectively, it depicts that consumers, once having broken a before-established rule, continue contradicting their standard because they feel like the situation is a lost cause (Herman et al., 1987). Applying this theory would adequately point out that once people have made an unhealthy purchase, they continue to do so. The authors disclosed that restrained eaters tended to consume more than unrestrained eaters, if they assume having shattered their diet goal (Polivy et al., 2010). Therefore, it appears that situations differ when consumers commence with an unhealthy choice.
6
therefore, eat more (Giner-Sorolla, 2001). Lastly, Zemack-Rugar, Bettman and Fitzsimons (2017) find that participants primed with guilt, behave less indulgent than participants who were primed with sadness. Seemingly, these studies support the idea of the adverse licensing effect.
In order to prevent consumers from taking part in the what-the-hell-effect and have them indulge in adverse licensing, shoppers could be confronted with the healthiness of a certain product to indirectly elicit feelings of guilt regarding past unhealthy choices. Consequently, they might be inclined to correct their negative self-image by buying healthy (Hofmann & Fisher, 2012). In line with this claim, Han, Duhachek and Agrawal (2014) advise marketers to use guilt or shame as mechanism in their advertising to remind people to live healthier. Consequently, implying guilt, whether through directly reminding customer of past actions or by passively reminding them of characteristics of healthy groceries, could be a way to intercept the buying of unhealthy products (Hofmann & Fisher, 2012). In brief, Table I gives an overview of the four theories provided in this review and their implications for choice-making.
Table I: Overview of Implications for Choice Making from Reviewed Literature Theoretical
Concept
First Choice Implications for Second Choice
Supporting Literature
Licensing Effect Healthy Unhealthy Khan & Dhar (2006)
Weibel, Messner, & Bruegger (2014)
Ishibashi, Miyazaki, & Katsutoshi (2015) Kivetz & Zheng (2006) Hui et al. (2009)
Goal Commitment Healthy Healthy Dhar & Simonson (1999)
Novemsky & Dhar (2005) Bargh, Gollwitzer, Lee-Chai, Barndolla, & Troetschel (2001)
What-the-hell-effect Unhealthy Unhealthy Polivy et al. (2010) Herman, Polivy & Esses (1987)
Adverse Licensing Effect
Unhealthy Healthy Hofman and Fisher (2012)
Giner-Sorolla (2001)
7
Supermarket Interventions
In order to achieve the before-discussed breakage of the two choice-making phenomena, in-store interventions deliver a great leeway. Therefore, the different forms of interventions and their effectiveness are discussed in the abstract-to-follow.
Health Interventions. Various studies have tested health interventions and their effectiveness in changing buying behavior (e.g. Adam & Jensen, 2016; Cameron et al., 2016), which lead to the establishment of meta-analyses concerning health interventions in supermarkets (e.g. Chandon & Wansink, 2012). In their review on food marketing, Chandon and Wansink (2012) give a detailed overview of the effectiveness of price, advertising, product and place interventions. According to the authors, pricing is a powerful tool that can increase the amount of healthy food purchased, especially within low-income households. It is however questionable how much of this consumption change is caused by the price discount and how much is due to a real understanding of the importance of a healthy life-style. Furthermore, they find that food and ingredients’ branding strongly influences the health perception of a product. It is therefore recommended to increase advertising for healthy items, to label products as healthy and highlight specific nutrition facts. Regarding the placing of products and location features, the authors demonstrate that the consumption of healthy food is a function of its accessibility and overall convenience. When healthy food is present or it is highly visible and accessible, its sales and intake increase. (Chandon & Wansink, 2012)
8
Taste Interventions. Neglected until this point but in fact also within the range of possible reactions of consumers is to show behavior which is opposing to the intention of an intervention, a concept acknowledged as reactance (e.g Clark, Goyder, Bissell, Blank, & Peters, 2007; Trump, 2016; Hanks, Just, & Wansink, 2014). Being reminded to consume healthy, consumers could therefore intentionally decide to consume unhealthy. Reasoning-wise, this effect occurs because individuals can experience an encroachment into their personal freedom when being confronted with interventions and consequently decide to ‘act out’ (Trump, 2016; Wendlandt & Schrader, 2007). Hence, it might be of interest to not only study an intervention that underlines the healthiness of a product but also one that stresses the tastiness of a grocery, which could possibly also lead to healthier choices caused by reactance. Forwood, Alexander, Hollands and Marteau (2013) show that a taste label increases the purchase of food items because of the improved tastiness perception. Furthermore, studies show that the higher the taste expectations, the higher purchase probabilities (Thunström & Nordström, 2015; Helfer & Shultz, 2014). Lastly, in a comparison of health and taste labels, Jacquot, Berthaud, Sghair, Diep and Brand (2013) find that taste labels are more effective in influencing taste expectations and thus, willingness to consume than health labels.
Endurance of Interventions. Reflectively, it was established that interventions have an immediate consequence on the choices they concern (Chandon & Wansink, 2012). However, it is questionable, to which extent the effect of those interventions remains throughout the shopping trip. In line with this consideration, Waterlander, Steenhuis, de Boer, Schuit and Seidell (2012) disclose that participants, who receive a discount on healthy food, purchase healthier at first but also acquire more calories overall, which demonstrates no long-lasting effect of the intervention. Therefore, health interventions’ short-term effectiveness might actually be offset by a lagged ineffectiveness. Furthermore, Cleeren et al. (2016) discover not only that on the short-term-perspective low fat choices increase food consumption but also that a significant long-term over-purchase result exists. Consequently, the long-term effects of interventions differ from their short-term effectiveness. To consider only the following choice of a health intervention would derive a limitation to this study.
9
shopping (Degeratu, Rangaswamy, & Wu, 2000). However, Wang, Minor and Wei (2011) discover that the different aesthetics of an online shopping environment influence consumer’ decision-making and thereby, their reception of store features, which could be transferred to store interventions. Furthermore, Kamakura (2012) also establishes an interdependency between purchases when analyzing online grocery shopping basket data. Lastly, regarding marketing interventions like prices, brand names and product information, labels as well as factual product information like calorie percentage becomes more important online. While the price sensitivity is higher in online stores, the combined effect of price and promotion activities is weaker (Degeratu, Rangaswamy, & Wu, 2000). Overall, it appears that, although the interdependence of choices is also present in online supermarkets, promotions could have different effects online.
10
Conceptual Model & Hypothesis
After having provided the theoretical basis concerning choice making, the hypothesis as well as conceptual model of this study are derived. In general, it is aimed to investigate how the levels of healthiness of choices affects the next choices made and if interventions can break the psychological effects that take place during this choice-making procedure. Furthermore, it is aimed to discover how much of the intervention effect remains for following choices and at the end of the shopping trip. To aid the understanding of the hypotheses, it is worthwhile mentioning that this paper entails a distinction into three experimental stages, which are applied in the development of the hypothesis and the remainder of this thesis. Overall, the three stages are defined as follows:
Stage 1 Effect of the healthiness of the previous grocery choice and the intervention on the healthiness of the current grocery choice.
Stage 2 Effect of the healthiness of the current grocery choice and the remaining effect of the intervention on the healthiness of the subsequent choice.
Stage 3 Effect of the intervention on the healthiness of the final shopping basket.
For a better understanding of the stages and the effect of the intervention, please refer to the visual representation of the conceptual model on page 12 of this thesis.
Hypotheses
Regarding the beginning stages (Stage 1 and 2) the different scenarios are displayed in Table II, accompanied by their respective underlying theory.
Table II: Overview of Scenarios Scenario
No. Previous Choice Intervention Current Choice Subsequent Choice Supporting Theory 1a (Control) Healthy None Unhealthy Healthy Licensing Effect
1b Healthy Taste Unhealthy Unhealthy (Stronger) Licensing
Effect
1c Healthy Health Healthy Healthy Goal Commitment
2a (Control) Unhealthy None Unhealthy Unhealthy What-the-hell-effect
2b Unhealthy Taste Unhealthy Unhealthy (Stronger)
What-the-hell-effect
2c Unhealthy Health Healthy Healthy Adverse Licensing
11
Stage 1: Previous and Current Choice. With respect to Table II, for customers who conduct a previous healthy choice, licensing would advocate their current choice to be unhealthy if no intervention is presented (Scenario 1a). However, if participants receive a taste intervention, this could intensify the licensing effect and motivate them to acquire unhealthier groceries in their choice to be followed (Scenario 1b). In case of encountering a health-motivating intervention, the licensing effect could be broken and consumers could be reminded to keep making healthy choices, resulting in goal-commitment (Scenario 1c). However, for customers that undertook an unhealthy previous choice, different scenarios disclose. For these participants, the what-the-hell-effect would propose that customers of an unhealthy previous choice, carry out another unhealthy choice, if their shopping scenario is not intervened (Scenario 2a). Moreover, literature implies that when customers of an initially unhealthy choice encounter an intervention for tasty food, they indulge in an even unhealthier food choice (Scenario 2b). Lastly, for a customer of an unhealthy previous decision, who encounters a health intervention, the current choice is predicted to be healthy, according to the adverse licensing effect (Scenario 2c). According to the listed set-ups, it is formally hypothesized:
H1a: The healthiness of the current choice is influenced by the healthiness of the previous choice.
H1b: The type of intervention influences the relationship in H1a. A health intervention leads to a healthy current choice while a taste intervention leads to an unhealthy current choice.
12
(Scenario 2b). Lastly, for a customer, that makes a healthy current choice and encounters a health intervention, the subsequent choice is predicted to be healthy, according to the adverse licensing effect (Scenario 2c). Therefore, it is formally hypothesized:
H2a: The healthiness of the subsequent choice is influenced by the healthiness of the current choice.
H2b: The type of intervention from Stage 1 remains influential for the relationship in 2a. A formerly-experienced health intervention leads to a healthy subsequent choice while an experienced taste intervention leads to an unhealthy subsequent choice. Stage 3: Remainder of the Shopping Trip. As aforementioned, the effect of most health interventions tested has, besides temporarily appealing to customers’ health-consciousness, when considered over the whole shopping trip, not led to a decrease in calories. Furthermore, it was established that grocery choices become unhealthier as the shopping trip continues. It appears that, although it is temporarily possible to disrupt the two psychological phenomena through an intervention, no long-term effect remains until the end of the shopping trip. Therefore, it is formally hypothesized that:
H3: The average amount of calories of the shopping basket does not differ between the intervention conditions.
After having defined the hypotheses, the conceptual model can be derived, see below. Conceptual Model
Figure I: Conceptual Model
13
and, while making this decision, encounters an intervention that is health-directed (underlining the healthiness of the product), taste-directed (emphasizing a great taste of the grocery) or not present. In Stage 2, the participant makes a subsequent choice with a certain measured degree of healthiness and in Stage 3, the participant continues his shopping trip with further purchases. At the end of the shopping trip, the average healthiness of the shopping basket is measured to assess whether the effect of the intervention persisted.
As first choices are usually driven by external factors, the medium timing of the intervention seems most feasible. Directly after entering the store, consumers may be affected by motives such as their diet plan or social influences. As the shopping trip continues, people become more depleted and exert less self-control (Hui, Inman, Huang, & Suher , 2013) which makes them more susceptible to marketing activities (Janssen, Fennis, Pruyn, & Vohs, 2008; Vohs & Faber, 2007). Logically, the intervention is not directly presented at the beginning of the shopping. Due to complexity, the effect of the past choices on the previous choice and the intervention are not evaluated in the course of this study, as well as the influence of all choices on the final healthiness of the shopping basket.
Methodology Research Design
Population & Sample Size. The study involved 360 participants from a panel of adult American consumers from Amazon Mechanical Turk, with an age range between 22 and 77 years. Participants were selected based on an age constraint (+18) to ensure that they made independent food decisions. As the utilized online supermarket contains American products as well as measurements, testing the interventions on American shoppers was most feasible. Furthermore, this panel contains participants that are representative for the American society, which ensures generalizability of the results (Amazon Mechanical Turk, 2018).
14
used for research purposes. This supermarket contains about 3,000 American products from 18 different product categories, ranging from animal food to household supplies and normal groceries. The store shows, except for the lack of actual payment transaction, all normal features of a regular online grocery store. When entering a category and selecting a product, the following information are displayed: name and image of the product, price, weight of the product and a product label containing nutrition information like calories, fat, saturated fat, sugar and salt. (van Ittersum, Wansink, Pennings, & Sheehan, 2013)
A quantitative assessment was chosen most appropriate for the research as customers are unique and it thus, needed to be established whether the results observed hold for a variety of individuals. Since the goal of the study was not to retain deep qualitative information, but to test the existence of an effect in a large and varied group of participants, this research type was found most appropriate. Secondly, it was opted for an experimental design because this did not only allow determining causality between treatments and their effects but it is also gave opportunity to control for variation and to randomly assign people to the treatment conditions. Furthermore, the laboratory experiment facilitated clear manipulation of the independent variable. Therefore, extraneous variables such as the presence of other people and knowledge of the store were minimized. (Aronson, Wilson, & Brewer, 1998)
Data Collection
Intervention Style. As discussed in the theoretical outlay, various possibilities exist for intervening shopping behavior of healthy food items like alternating the price, the arrangement of the products or adding a product label. However, a price intervention does not deem to increase the health awareness of customers but only appeals to their price interest, which is not an effective long-term method to change buying behavior and thus, decrease obesity (Cameron et al., 2016). Due to the online feature of this study, a change of the product itself or their arrangement to an eye-level was not feasible. Therefore, the chosen intervention style was a product label, which was also proven effective in various other studies (e.g. Surkan et al., 2016; Cameron et al., 2016; Teisl & Levy, 1997).
15
product. Thirdly, Hui et al. (2009) state that yogurt accounts for 11.5% of buying which makes it a familiar product but yet leaves enough room for testing an intervention as it is not the most common product sold. Lastly, yogurt sales are increasing and make up an important share of American grocery shopping (Statista, 2018a), which ensures their relevance for consumers.
Measures. Having reflected the different possibilities to define healthiness in the theoretical foundation, it was, as obesity is mainly driven by the general over-consumption of calories (Wright & Aronne, 2012), decided to measure healthiness in this study by the amount of calories that a piece of grocery entails, in relation to the average of its category. The healthiness of a piece of grocery was therefore determined by subtracting the calories of a product (product_calories.n) from its category average (average_calories_ for_category.n). If the value was equal to or above zero, the grocery was considered healthy. A score below zero (thus, the product has more calories than the category average) was categorized as unhealthy. For instance, if a participant decided to buy a yogurt containing 45kcal/100g while the category had an average of 64 kcal/100g, the difference between the two was 19, a positive value; therefore this yogurt was ‘healthy’. If, on the other hand, participants bought a pizza with 60kcal/100g compared to a category average of 39kcal/100g, this resulted in a negative value (-21); this piece of grocery was unhealthy.
16
study was pretested with 20 participants. All participants agreed that the study was interesting, understandable and easy to follow, which allowed continuance with the actual study.
Procedure. Participants were presented with a cover story of taking part in a normal grocery-shopping trip. They received a shopping list, which contained ten grocery categories, and clear shopping instructions (see Appendix A). The groceries were randomized across participants with exception for the product category connected to the intervention, yogurt. The grocery yogurt was kept stable at the fifth position on the shopping list to ensure that all participants met the intervention in the same mental state. The concrete choices the customer made were not intervened to ensure simulating a real shopping experience. After having made three choices, the customer made a fourth choice (the previous choice in Figure 1) which was either high or low in amount of calories and respectively healthy or unhealthy. Consequently, the customer made a fifth grocery choice (yogurt, denoted as current grocery choice in Figure 1) according to his shopping list, which was also relatively healthy or unhealthy. With this fifth choice, the customer encountered either (1) no intervention, (2) the health label on a yogurt low in calories or (3) the taste label on a yogurt higher in calories (see Figure II). The product with the intervention was presented in the first row of the product category to ensure visibility; the order of the yogurts in the store as well as the prices and other store features were kept stable across all participants. After having made the sixth choice (denoted as subsequent choice in Figure I) the customer continued with making his last four choices. After shopping, participants checked out, returned to the survey and answered several demographic and health and shopping related questions (see Appendix A).
Figure II: Interventions in the Online Supermarket
17
they faced questions regarding their frequency of their (online) shopping behavior with estimates based on the average behavior of American citizens (Statista, 2018b; Statista, 2018c; Statista, 2018d). Furthermore, height and weight data were collected to calculate participant’s BMI. Their attitude towards and motivation for healthy eating was assessed (Dijkstra, Neter, Brouwer, Huisman, & Visser, 2014; Astrosm & Rise, 2001). It was also assessed that participants had no nutritional restrictions and if they were currently dieting (Emerton, 2007; Dinu, Abbate, Gensini, Casini, & Sofi, 2016; Just, Heiman, & Zilberman, 2007). Moreover, they faced questions regarding the credibility and realism of the online store according to similar study approaches (e.g. van Ittersum et al., 2013). Lastly, the demographic questions concerned the participant’s age, gender and profession to be able to profile the participants. For the complete questionnaire, please refer to Appendix A.
Results General Analysis
Descriptive Analysis. To obtain a general sense of the data, a descriptive analysis was undertaken (see Appendix C and Table III), analyzing the demographic features and shopping behavior of the population in question. It was ensured that participants were responsible for their own household (Age ≥ 18) and the main shopper in the family. Participants who did not fulfill those criteria were excluded from further participation.
18
Table III: Descriptive Statistics (N=352)
Descriptive Statistics
N Minimum Maximum Mean Std.
Deviation Age in Years 352 22 77 38.05 10.78 Income in Dollar ** 352 0 80000 5030.78 9314.69 Expenditure Groceries in Dollar* 352 0 500 97.73 59.03 Shopping Frequency* 352 0 7 1.32 0.75 Online Shopping Frequency** 352 0 10 1.47 2.05 BMI (Calculated) 352 0 63 26.71 7.20 * Per week, **Per month
While comparing the descriptive features with the average American citizen, analysis showed representativeness of the data. Participant’s monthly income as well as shopping behavior for groceries and online shopping were in accordance with American average values (Statista, 2018b; Statista, 2018c; Statista, 2018d: OECD, 2016). In addition, the average BMI score met the American average of 26.5 for woman and 26.6 for men, displaying that the average American is slightly overweight (U.S. Department of Health and Human Services, 2000). Although the gender split was slighltly too male dominated to be perfectly in accordance with American standards (Statista, 2018e), it was established that the data is representative of the American society and results can be used for quantification purposes.
19
Table IV: Summary Factor Analysis (N=352)
Factor Loading
Item 1 2
Factor 1: Importance of Healthiness (FACHealth) (a = .925)
(1) The healthiness of food has little impact on my food choices. .847 n.A. (2) I am very particular about the healthiness of food. .903 n.A. (3) I eat what I like and I do not worry much about the healthiness of food. .847 n.A. (4) It is important to me that my daily diet is low in fat. .614 n.A.
(5) I always follow a healthy and balanced diet. .792 n.A.
(6) It is important for me that my daily diet contains a lot of vitamins and
minerals. .817 n.A.
(7) The healthiness of snacks make no difference to me. .871 n.A. (8) I do not avoid foods, even if they may raise my cholesterol. .787 n.A.
Mean Score (out of) 4,31 (7,00)
Factor 2: Importance of Tastiness (FACTaste) (a = .699)
(1) I do not believe that food should always be source of pleasure. .617 .251 (2) The appearance of food makes no difference to me. .136 .720 (3) When I eat, I concentrate on enjoying the taste of food. .742 .159 (4) It is important for me to eat delicious food on weekdays as well as on
weekends. .807 .166
(5) An essential part of my weekend is eating delicious food. .870 -.510 (6) I finish my meal even when I do not like the taste of a food. .099 .796
Mean Score (out of) 4,73 (7,00) 4,73 (7,00)
Factor 3: Reason for Healthy Diet (FACReas) (a = .842)
(1) Because I feel that I want to take responsibility for my own health. .810 -.006 (2) Because I would feel guilty or ashamed of myself if I did not eat a
healthy diet. .331 .605
(3) Because I personally believe it is the best thing for my health. .816 -.041 (4) Because others would be upset with me if I did not. -.034 .813
(5) I really don't think about it. .750 -.044
(6) Because I have carefully thought about it and believe it is very important
for many aspects of my life. .808 .007
(7) Because I would feel bad about myself if I did not eat a healthy diet. (7) .526 .488 (8) Because it is an important choice I really want to make. (8) .866 .026
(9) Because I feel pressure from others to do so. -.133 .827
(10) Because it is easier to do what I am told than to think about it. .134 .742
(11) Because it is consistent with my life goals. .821 0.13
(12) Because I want others to approve me. -.044 .837
(13) Because it is very important for being as healthy as possible .839 -.72
(14) Because I want others to see I can do it. .077 .739
(15) I don't really know why. .686 -.214
20
Table IV (continued): Summary Factor Analysis (N=352) Factor 4: Realism of Online Store (FACReal) (a = .658)**
(1) The prices presented during the grocery trip were realistic. .042 .765 (2) I was familiar with the products presented during the grocery trip. -.003 .780
(3) I liked this grocery trip. .402 .568
(4) I liked the way the nutritional information was presented during this
grocery trip. .934 .084
(5) I understood the nutritional information format that was presented during
this grocery trip. .901 .065
Mean Score (out of) 5,82 (7,00) 4,65 (7,00)
Factor 5: Attention (FACAtten) (a = .502)**
(1) Product Specifications .811 n.A.
(3) Nutritional Information (Calories, Fat, Saturated Fat, Sugar, Sodium) .806 n.A.
Mean Score (out of) 3,91 (7,00)
Factor 6: Price Sensitivity (FACPriceSens) (a =.502)
(1) For me, price is decisive when I am buying a product. .889 n.A. (2) Price is important to me when I choose a product. .875 n.A. (3) I generally strive to buy products at the lowest price. .857 n.A.
Mean Score (out of) 5,48 (7,00)
Factor 7: Mood (FACMood) (a = .891)*
(1) Bad- Good .926 .245 (2) Sad- Happy .925 .291 (3) Displeased- Pleased .927 .248 (4) Calm- Excited .136 .865 (5) Tired – Energetic .537 .695 (6) Sedated – Aroused .272 .814
Mean Score (out of) 6,92 (9,00) 5,47 (9,00)
Note: Factor loadings in bold display factor membership. N.A.: Not applicable (no second factor)
*The price item (Item 2) was not included in the factor but kept separate due to a low loading (.201) **Loadings after rotation
21
reasons). With a score of 5.34, participants scored much higher on personal importance than on outside pressure (Score of 2.60). Seemingly, participants were rather motivated to follow a healthy diet to keep up a good personal health than to impress their peers. The fourth factor analysis concerned the realism of the grocery store and extracted two factors, FACRealGen and FACRealNutri. FACRealGen combined items that expressed a general liking and realism of the store (general realism) and FACRealNutri described the understanding and liking of product information and specifications (nutrition realism). Respondents scored 5.82 and 4.65 on the factors in question, showing that they assessed their shopping experience as realistic.
Regarding the attention span of respondents, FACAttenPro expressed how much attention participants paid to general characteristic of products and the store and FACAttenPri displayed how much attention participants paid to prices. The middling scores for these factors were 3.91 and 5.55. Thus, participants paid, on average, more attention to prices than to product characteristics and nutritional information. In addition, price sensitivity was expressed through one factor, titled FACPriSen on which participants scored 5.48, which again underlined participants’ price-awareness. Lastly, participants’ mood was expressed through two factors, FACMoodGe and FACMoodInt. The first comprised general and less intense feelings (general mood) while the latter combined items that concerned powerful and extreme emotions (intense mood). On average, participants scored 6.92 on the first and 5.47 on the second factor, which resolved in generally positively-excited mood.
All factors met the criteria for factor analysis (KMO > 0.5; p-value < 0.05 and communalities > 0.4) and were reliable according to Cronbach’s Alpha with a value above the respective threshold (see Table IV). The factors were thus deemed adequate for further usage in the following analysis.
22
Although the shopping instructions clearly advised participants to purchase yogurt as their fifth choice, ensuring that they encounter the intervention at the same time, data inspection showed that participants did not always follow this mandate (see Appendix E). To account for this limitation, instead of analyzing participant’s fifth choice, it was inspected which purchase was the yogurt and this choice was considered participants’ current choice. The choice before the yogurt was, in accordance with the study model, denoted as previous choice and the purchase decision after the yogurt was considered the subsequent choice. To control for the effect of the yogurt decision being different across participants, a covariate variable called YogurtPosition was built and tested, indicating at which position participants acquired the yogurt and thus, encountered the intervention. For each choice, respective variables were created (see Table V).
Table V: Choice Variables
Name of Choice Name of Variable Description
Previous Choice PreChoiceCalDif Calorie difference between the choice and its category average for the choice before the yogurt (and intervention)
Current Choice ChoiceCalDif Calorie difference between the choice and its category average for the yogurt choice (and intervention)
Subsequent Choice
PostChoiceCalDif Calorie difference between the choice and its category average for the choice after the yogurt (and intervention)
For participants, who did not undertake any of the above indicated choices, data was marked missing and was deleted pairwise. This is evaluated an appropriate method to deal with missing data (Lewis-Beck & Bryman, 2007).
Results Stage 1
23
For this purpose, three dummy variables were created as the manipulation check had four levels (0= Saw no label 1=Saw the health label 2= Saw the taste label 3=Saw both labels). The respective dummy variables were denoted DumManT, DumManH and DumManB.
As variable inclusion method for multiple regression, ‘specific to general’ was dissected, which evolved by first including the main effect, then the interaction effects and lastly, the covariates. It was opted for this route as, based on aforementioned literature, the label as well as the previous choices were expected to have significant and possible interaction effects. Therefore, various effects should be tested and it needed to be assessed whether other covariates influenced the outcome. Consequently, the following equations should be judged by multiple regression.
Specific Model: XChoiceCalDif= β1*XPreChoiceCalDif + β2*YDumConTaste+ β3*YDumConHealth
General Model: XChoiceCalDif= β1*XPreChoiceCalDif + β2*YDumConTaste+ β3*YDumConHealth + β4*(XPreChoiceCalDif
*YDumConTaste) + β5*(XPreChoiceCalDif *YDumConHealth)
24
The results of the specific model are displayed in Table VI and more extensively in Appendix F. In this model, the R Square accounted to 2% and the adjusted R Square to 1.1%, indicating that this specific model had no good fit. However, a high variance in the dependent variable seemed logical when considering that consumer choices are unique and difficult to estimate (e.g. von Fiore, La Sala, Gallo, & Tsoukatos, 2017). In consideration of Table VI, PreChoiceCalDif predicted calories of the current choice (Beta= .100; p= .019) and the unstandardized coefficient indicated that PreChoiceCalDif had a positive effect on current calorie difference. This denouement expresses that, for a previous healthy choice, the current choice was healthy. For instance, a consumer made a healthy previous choice with a calorie difference of 20 calories less than the product category average. The coefficient would then indicate that his current choice has 20*0.100= 2 calories (XChoiceCalDif= 0,100*XPreChoiceCalDif) and
is therefore healthy, instead of unhealthy, as would have been predicted by the licensing effect. Consequently, proof for breaking the licensing effect was found. On the other hand, when considering those consumer who started with an unhealthy choice, their next choice was still unhealthy (from a calorie difference of -20 to 0.100*-20= -2 calories). Therefore, proof for the existence of the what-the-hell-effect was found, but not for its breakage. It remains to further analysis in the following abstracts to dissect what broke the licensing effect.
25
Table VI: Regression Results Stage 1
General Model. For the development of a more general model, the interaction effects between the label intervention and the pre-yogurt-choice calorie differences were included (see Table VI). The underlying purpose was to analyze whether effects only existed in certain conditions. Likewise, the assumptions for multiple regression had to be re-assessed (Appendix G). Since the sample size and dependent variable were identical to the specific model, the size was adequate but normal distribution was not given but accepted to be ignorable. Moreover, multicollinearity was not problematic, a linear relationship between the independent and the dependent variable was found and homoscedasticity could be rejected. Thus, multiple regression was appointed as adequate research method.
Considering an R Square of 2.6% but an adjusted R Square of 1.1%, the model fit did not change in comparison to the specific model and was still not high (see Appendix G). Considering the included interactions, none of them reached a significant level (see Table VI). Furthermore, the significant effect from the specific model did not hold in this model and also the conditions remained insignificant. Thus, in this model, neither the previous choice nor the condition or interaction effects affected the outcome variable. The breakage of the licensing effect as well as the missing disruption of the what-the-hell-effect faded in this model.
Parameter Sig. SE Parameter Sig. SE Parameter Sig. SE
Main Effects PreChoiceCalDif .100 ** .019 .043 .035 .637 .074 .022 .764 .073 DumConHealth 4.003 .349 4.265 3.844 .368 4.266 4.087 .347 4.339 DumConTaste 4.093 .344 4.323 4.049 .350 4.325 5.273 .232 4.401 Interaction Effects InteractPreHealthL .143 .167 .103 .163 * .091 .103 InteractPreTasteL .051 .631 .105 .057 .583 .104 Covariates DumManH 6,598 .227 5.448 DumManT -16.458 * .075 9.212 DumManB -1.322 .863 7.645 FACHealth 5.583 *** .003 1.847 BMI -.001 .998 .254 Gender .387 .913 3.534 Diet 3.100 .459 4.183 Positon -1.200 .677 2.873 Constant -16.413 *** .000 3.035 -16.404 *** .000 3.035 -12.409 .489 17.917 Number of individuals
Specific model General Model General Model
with Covariates
336 336 336
26
Covariates. To assess whether some of the factors built in the aforementioned factor analysis, the manipulation dummy and variables from the descriptive analysis affected the outcome variable, important factors and descriptive variables were included (Appendix H). The assessed variables were the manipulation dummies, the main effects of the health interest factor and the variables BMI, gender and diet status. In the same way, it was assessed whether the position of the intervention had an effect. Other covariates were not analyzed, as it would have exceeded the scope of this study. However, the factors taken into consideration are common factors studied and were thus deemed sufficient (e.g. Papies et al., 2014). Regarding the assumptions for multiple regression, all requirements besides normal distribution of the dependent variable could be confirmed. The abnormal distribution was neglected, with identical reasoning as to the specific model.
Overall, the model fit increased to an R Square of 7.5% and adjusted R of 3.7%, which expressed an improvement. Surprisingly, the main effect of the position of yogurt was not significant; it thus did not make a difference at which point in the shopping trip participants bought the yogurt (Beta= -1.200; p= .677). However, the variable InteractPreHealthL reached marginal significance (Beta= .163; p= .091). This variable constituted the interaction effect between the health label and the pre-yogurt choice. With a parameter of 0.163, this effect expressed that participants, who were in the health condition and made an unhealthy previous choice, made an unhealthy current choice. For participants with a healthy previous choice however, the health label led them to another healthy choice. Similar to the example in the specific model (XChoiceCalDif= 0,163*(XPreChoiceCalDif *YDumConHealth)) a healthy pre-choice of 20
27
Figure III: Interaction effect Stage 1
Furthermore, a significant effect was found for the health interest factor, FACHealth (Beta= 5.583; p= .003), indicating that participant’s health interest had an essential consequence for the amount of calories of the current choice. The higher the health interest of a person, the less calories that participant purchased in the current choice. Lastly, although it seemed in first instance that encountering a taste label did not affect the outcome variable, when inspecting the manipulation check a marginally significant effect occured for the taste condition (Beta= -16.458; p= .075). Therefore, those participants who consciously noticed the taste label, made an unhealthier current choice, which was proof for a stronger licensing effect, caused by the label.
28
Table VII: Results Stage 1
Previous Choice Significant
Variable Coefficient Implications for Current Choice
Healthy/ Unhealthy Taste Label -16.458 Unhealthier than Previous
Choice
Healthy Health Label*
Previous Choice
0.163 Healthy
Unhealthy Health Label*
Previous Choice
0.163 Unhealthy
Healthy/ Unhealthy Health Interest 5.583 Healthier than Previous
Choice
Considering hypotheses H1a, H1b and the results in Table VII, the data gathered suggested that hypothesis H1a had to be rejected while hypothesis H1b was partly confirmed. Although in the specific model previous choices affected current choices, this effect did not hold in the general model, rejecting hypothesis 1a. However, for participants in the healthy label condition, the label affected their current choice in the final model, partly proving hypothesis 1b. For participants in the tasty condition, the label had a main effect if they consciously saw it but did not moderate the relationship between the dependent and independent variable. With these findings in mind, it was established that the licensing effect could be broken, but only for those participants who made a healthy previous choice and encountered a healthy label. Lastly, the covariate health interest influenced the calories of the current choice.
Results Stage 2
Method. In this stage, the assumptions under discussion were hypotheses H2a and H2b, which considered the effect of a current decision on the subsequent decision and a possible moderation effect of the before-experienced intervention. Based on the conceptual model, the subsequent choice was treated as dependent variable, while the current choice and the intervention type served as independent variables. Identical to procedures in Stage 1, the dummy variables were utilized to examine the effect of the intervention and it was controlled for the manipulation check. Also similar to Stage 1, the method of ‘specific to general’ was determined as modus operandi with the general model complementing the specific models by the interaction effects. Subsequently, the following regression equations emerged:
Specific Model: XPostChoiceCalDif= β1*XChoiceCalDif + β2*YDumConTaste+ β3*YDumConHeallth
General Model: XPostChoiceCalDif= β1*XChoiceCalDif + β2*YDumConTaste+ β3*YDumConHeallth + β4*(XChoiceCalDif
29
Specific Model. To lay the foundation for multiple regression, several assumptions were assessed (see Appendix I). Due to consideration of the identical data set, the sample size exceeded the critical level. Furthermore, the dependent variable PostChoiceCalDif exhibited no normal distribution with a Shapiro-Wilk and Kolmogorov-Smirnov significance value of p=.000. However, the visual inspection of the histogram evicted normally distributed data. As the sample size was substantial, the one existing outlier was not abnormal and multiple regression could be considered stable towards abnormal distribution (Li, Wong, Lamoureux, & Wong, 2012) the analysis was continued. Subsequently, it was ensured that no multicollinearity issues occurred with VIF-Scores below the cut-off value. Considering the linear relationship between the independent variable and the dependent variable, the scatterplot showed an almost linear distribution, meeting this requirement. Regarding homoscedasticity, the standard residual was within the adequate range (Min.: -2.925, Max.: 2.722) and the visual inspection of the scatterplot evicted an acceptable distribution. Hence, multiple regression was assessed appropriate and the analysis could be commenced.
30
Table VIII: Regression Results Stage 2 (N=336)
General Model. To further investigate the findings, the interaction between the intervention and the current choice calorie difference were included for the general model. It was aimed to test if outcomes differed when considering interaction effects. For the purpose of this new model, the assumptions for multiple regression had to be re-assessed. As shown in Appendix J, sample size and dependent variable results were identical to the beforehand executed analysis. Multicollinearity was not problematic and there was clear evidence for a linear relationship between the independent and the dependent variable. Despite the residual minimum being slightly out of line, homoscedasticity was rejected after visual inspection and adequate distribution of data in the scatterplot. In line with these requirements, multiple regression could be undertaken.
The outcomes of multiple regression suggested an increase in the model fit to an R Square of 5.2% and an adjusted R Square of 3.8%. A glance at Table VIII reveals that the significant main effect of the current choice held across models (Beta= -.36; p= .006 and that the health label reached significance, after inclusion of the interaction effects. Furthermore, a marginally significant effect was found for the health label condition (Beta= 11.384; p= .086). This significant effect demonstrated that participants, who consciously encountered a
Parameter Sig. SE Parameter Sig. SE Parameter Sig. SE
Main Effects ChoiceCalDif -.227 *** .004 .078 -.36 *** .006 .131 -.364 *** .007 .135 DumConHealth 5.325 .385 6.126 11.384 * .086 6.615 12.297 * .072 6.820 DumConTaste 5.642 .364 6.207 5.408 .421 6.706 5.237 .447 6,881 Interaction Effects InteractChoHealthL .455 ** .016 .187 .476 ** .014 .193 InteractChoTasteL -.055 .772 .189 -.056 .773 .193 Covariates DumManH -2.652 .741 8,006 DumManT 7.667 .57 13.482 DumManB -8.068 .468 11.115 FACHealth 1.346 .621 2.722 BMI .202 .586 .370 Gender -2.886 .576 5.153 Diet -1.060 .861 6.069 Positon 3.176 .461 4.304 Constant -6.523 .151 4.528 -8.65 * .72 4.793 -25.192 .335 26.104 Number of individuals
Specific model General Model with CovariatesGeneral Model
336 336 336
31
healthy label, bought healthier than participants who did not. Furthermore, a marginally significant interaction effect was found for the interaction between the current choice and the health label (Beta= .455; p= .016). This expresses that participants of a healthy current (yogurt) choice and the health label, made another healthy subsequent choice, indicating that the licensing effect was disrupted. However, for those participants whose yogurt choice was unhealthy, the subsequent choice was also unhealthy, indicating that the what-the-hell-effect could was not broken by the health intervention, in contrast with the results from the specific model. No significant effects were found for the taste label.
Covariates. On these grounds, covariates were included forwardly to administer if any variables enhanced model fit. Identical to Stage 1, it was concentrated on the manipulation dummy, the most important demographic variables (BMI, gender, diet status) and the health interest factor. Furthermore, it was assessed whether the position of the intervention influenced the dependent variable. Before commencing, the assumptions for multiple regression could be confirmed, except for the requirement of normal distributions and homoscedasticity. However, normal distribution was, in line with the same argumentation as in the other studies, neglected and homoscedasticity was not considered an issue after visual inspection of the scatterplot (see Appendix K).
32
Figure IV: Interaction Effect Stage 2
Consequently, it can be concluded that the licensing effect was broken in the healthy label condition, as highlighted in green in Table IX, the what-the-hell-effect, however, could not be disrupted. In a separate test, it was assessed whether three-way interaction effects between the label conditions, the current choice and the manipulation check were significant. However, no significant effects were found. Although it appeared in first instance that the licensing effect was not broken but the what-the-hell-effect was, the consideration of the final outcomes leads to a different conclusion, in line with the results from stage 1.
Table IX: Results Stage 2
Current Choice Significant Variable Coefficient Implications for Subsequent Choice
Healthy Calorie Difference Current
Choice
-.364 Unhealthy
Unhealthy Calorie Difference Current
Choice -.364 Healthy
Healthy/Unhealthy Health Label 12.297 Healthier than Current
Choice
Healthy Health Label*Current Choice .476 Healthy
Unhealthy Health Label*Current Choice .476 Unhealthy
33
The taste label, however, did not cause unhealthier subsequent choices, leading only to a partial support of hypothesis 2b.
Results Stage 3
Method. Finally, the effect of the intervention on the overall calories of the shopping basket was assessed, investigating hypothesis 3. The amount of overall calories after the shopping trip served as dependent variable, CalBasket, while the intervention dummies served as independent variables. Although the interventions did not show significant effects in individual choices in Stage 1 and 2, it was aimed to analyze if the labels affected the overall amount of calories purchased by respondents. Identical to the previous experiment stages, the method of ‘specific to general’ was determined and the conceptual model was reflected in the following regression equation:
Specific Model: XCalBasket= β1*YDumConHealth + β2*YDumConTaste
Specific Model. To ensure the data in question was appropriate for multiple regression analysis, several assumptions were assessed (see Appendix L). Firstly, the sample size was evaluated adequate and the dependent variable CalBasket was normally distributed with a Shapiro-Wilk value of p=.200, Kolmogorov-Smirnov value of p=.106 and a satisfactory embodiment in the histogram. The two outliers were closely inspected and, besides a comparatively low calorie amount, no abnormalities were observed so they were incorporated in the further analysis. Furthermore, the VIF-Scores were below the critical value, which advised that multi-collinearity was not problematic. Moreover, considering the linear relationship between the independent variable and the dependent variable, the points in the probability plot followed a clear linear line, therefore, also this requirement was confirmed. The last assumption, homoscedasticity, was also confirmed after visual inspection of the scatterplot and residual values within line, which evicted a rectangular distribution of data. Therefore, all assumption of multiple regression were approved.
34
Table X: Regression Results Stage 3
Specific model General Model
with Covariates Prm. Sig. SE Prm. Sig. SE Main Effects DumConHealth -17.76 .349 18.922 -24.281 .208 19.259 DumConTaste -8.024 .675 19.097 -19.176 .326 19.480 Covariates DumManT -7.365 .758 23.847 DumManH 51.168 .205 42.722 DumManB 23.152 .489 33.408 FACHealth -20.292 ** .014 8.173 BMI -.204 .858 1.137 Gender 28.198 * .072 15.604 Diet -23.080 .213 18.502 Position -6.728 .600 12.815 Constant 1174.24 *** .000 13.441 1183.12 *** .000 79.081 Number of individuals 327 327 *** p<0.01. ** p<0.05. * p<0.1
Covariates. The analysis was continued by scrutinizing the significance of covariates to determine if they affected the amount of calories in the final shopping basket. For this purpose, the same covariates as in Stage 1 and 2 were investigated. In this model, all requirements of multiple regression were met.
The respective model had an R Square of 5% and an adjusted R Square of 2%, which showed an increase in model fit (see Appendix M). A closer look at the data indicated that the covariates of health interest (Beta= -20.292; p= .014) and gender (Beta= 28.198; p= .072) were significant. Male participants bought on average more calories and a high health interest led to less calories in the final shopping basket, as shown by the parameter. Moreover, the main effect of the position of yogurt had no effect; it did therefore not make a difference at which point in the shopping trip participants bought the yogurt. Also in this model, the independent variables from the specific model remained insignificant and thus the intervention did not affect calorie values in general. (see Table X)
