THE EFFECT OF AN INTERVENTION ON HEALTHY-SHOPPING DYNAMICS What is the impact of a healthy or tasty intervention on the relative healthiness of subsequent
food choices?
18 June 2018
THE EFFECT OF AN INTERVENTION ON HEALTHY-SHOPPING DYNAMICS
What is the impact of a healthy or tasty intervention on the relative healthiness of subsequent food choices?
Master Thesis, MSc Marketing department
University of Groningen, Faculty of Economics and Business
18 June 2018 Laura Klinkenberg Studentnumber: S2370654 Bloemstraat 8A 9712LD, Groningen Tel: 0639420047 E-mail: l.m.klinkenberg@student.rug.nl Supervisor:
Prof. dr. ir. Koert van Ittersum
Second supervisor: Martine van der Heide
Examiner:
MANAGEMENT SUMMARY
Approximately 20 years ago, obesity reached the status of ‘epidemic’ (WHO, 2018). Currently, obesity is still one of the biggest health issues worldwide and researchers even forecast a 130% increase by 2030 (Finkelstein et al., 2005). Moreover, overweight and obesity carry several alarming consequences such as various health risks, both physical and psychological (Hill, Wyatt, Reed, & Peters, 2003; Ng et al., 2013). Furthermore, overweight and obesity also have an excessive impact on public health care costs (Finkelstein, Ruhm & Kosa, 2005).
An important driver for the increase in obesity seems to come from developments in the global food system, such as more processed, economical, and energy-dense foods (Asfaw, 2011; Swinburn et al., 2011). In addition, the effective and better targeted marketing also plays a role in the overconsumption of unhealthy foods (Swinburn et al., 2011). As a consequence, food marketers have been repeatedly criticized and appointed as the leading cause of the increase in obesity (Chandon & Wansink, 2012).
Though, food marketers and practitioners could also have a positive impact on consumers’ health and well-being. Hill et al. (2003) found that an intervention that leads to a decreased energy intake of just 100 kilocalories per day could cancel out weight gain in approximately 90% of the population. More importantly, a decrease of 100 kilocalories per day should be achievable by merely choosing a relatively healthier alternative in the same product category. For this reason, interventions with the goal to promote healthier food choices warrant increased attention.
However, previous research on interventions has mainly been focused on the impact of an intervention on single product purchases. Nevertheless, consumers in the real world often need to make decisions after they already made other choices, leading to a series of consecutive choices (Khan & Dhar, 2006). Therefore, existing knowledge, based on single choices or single product purchases, might be inadequate in examining the effectiveness of interventions. Also, to the best of my knowledge, no research yet has focused on the effect of an intervention on shopping trip dynamics.
the subsequent choice(s) consumers make? Also, does the type of intervention (healthy or tasty) influence the effect of the intervention?
Furthermore, the data that is used to test the research questions of the current study is derived from an online questionnaire that included the simulation of a grocery shopping environment and a survey. The participants were divided over eight different conditions, depending on the intervention they received (control, neutral, healthy, and tasty) and the timing of the intervention (early or late). Subsequently, the participants were asked to shop for 16 food products within 16 product categories, of which one was the cookie category (i.e. intervention category).
Next, the data was analyzed by means of hybrid panel regressions and multiple analyses of variance (ANOVA). The results of the hybrid panel regressions showed that a relatively healthy previous choice, leads to a subsequent unhealthier choice. On the contrary, a relatively unhealthy previous choice, leads to a subsequent healthier choice. However, the current study did not find any significant effects regarding the introduction of a new cookie on the relative healthiness of food choices. Similarly, several analyses of variance (ANOVA) showed there were also no significant differences between the group means for the different types of interventions and the different timing conditions.
PREFACE
My interest in this topic stems from my educational background in both Psychology and Marketing. I have always been passionate about consumer behaviour and in specific, consumer behaviour within the food retailing industry. Moreover, this topic also sparked my interest because I think it is very relevant. Currently, we are still facing a worldwide obesity problem. I believe marketing plays a great role in consumer behaviour and consequently the food purchasing behaviour of consumers. This influence can be in disadvantage of consumers’ health, however I believe this influence could also be used to positively affect consumers’ well-being. Accordingly, more insights in the food purchasing behaviour of consumers are needed to be able to nudge or steer consumers in the right direction.
I would like to thank Prof. dr. ir. Van Ittersum, professor Marketing and Consumer Well-Being, for his guidance and supervision throughout this entire period. Additionally, I would like to thank my second supervisor, promovendus van der Heide, for always being willing to help me when confronted with problems, especially during the data preparation and analysis phase.
I hope you will enjoy reading this thesis.
Laura Klinkenberg
TABLE OF CONTENTS
Title page ii
Management summary iii
Preface v Table of contents vi
1. Introduction
1
2. Theoretical background
4
2.1 Licensing behaviour 5 2.2 Goal commitment 6 2.3 Reparative behaviour 6 2.4 ‘What-the-hell effect’ 6 2.5 Impact of interventions 8 2.6 Type of intervention 92.7 Timing of the intervention 10
2.8 Conceptual model 11
3. Research methods
11
3.1 Design and research type 11
3.2 Participants 12 3.3 Procedure 12 3.4 Stimuli 14 3.4.1 Prize package 14 3.4.2 Food products 14 3.4.3 Intervention 15 3.4.4 Timing 16 3.4.5 Psychographics 16 3.5 Measurement of variables 16 3.5.1 Relative healthiness 16
3.5.2 Lagged relative healthiness 17
3.5.3 Timing and type of intervention 17
3.5.4 Control variables 18
3.6 Plan of analysis 19
4. Results
20
4.1 Reliability analysis health consciousness 21
4.3 Correlations 23
4.4 Hybrid panel data 23
4.5 Model building 24
4.6 Model estimation 24
4.6.1 Basic model 24
4.6.2 Intervention dummy model 25
4.6.3 Lagged intervention dummy model 25
4.6.4 Two lags intervention dummy model 26
4.6.5 Intervention and lagged intervention dummy model 27
4.6.6 Healthy and tasty dummy model 27
4.7 Robustness check 29 4.8 ANOVA 30 4.8.1 Results 31 4.9 Hypotheses 33
5. Discussion
35
5.1 Theoretical contributions 36 5.2 Practical contributions 36 5.3 Limitations 37 5.4 Future research 386. Conclusion
39
References
40
Appendices
45
A. Complete questionnaire of the control condition 45
B. Introduction of new cookie in the intervention conditions 53
C. G* power analysis 54
D. Reliability analysis health consciousness 55
E. Kolmonogrov-Smirnov and Shapiro-Wilk ANOVA 56
F. Q-Q plots ANOVA 57
G. Levene’s test and Welch ANOVA 58
1. INTRODUCTION
Globally, the prevalence of overweight and obesity is high and has been increasing for four decades now (Livingston, 2018; NCD, 2016; Ng et al., 2013). In 2016, approximately 39% of adults were overweight, defined by the World Health Organization (WHO) as a body-mass index (BMI) of between 25 and 30 kg/m², and around 13% of adults were obese, defined as a BMI of 30 kg/m² and above (WHO, 2018). The trend in elevated BMI is not only noticeable among adults, but also among children and adolescents. Accordingly, WHO estimated that in 2016 approximately 18% of the children and adolescents were overweight or obese as opposed to just 4% in 1975 (WHO, 2018). An important driver for the increase in obesity seems to come from developments in the global food system, such as more processed, economical, energy-dense and effectively marketed food (Swinburn et al., 2011). Besides changes in the food system, there is also a reduction in physical activity (Livingston, 2018). In other words, there is an increase in intake of energy-dense foods while, at the same time, there is a decrease in physical activity. This leads to an energy imbalance between calories that are consumed (dietary energy intake) and calories that are dispensed (energy output), which eventually causes people to gain weight.
Moreover, overweight and obesity bear various health risks. Several large pooling studies used for the Global Burden of Disease study by The Lancet in 2013, found persistent risks as BMI rises above 23, specifically for cardiovascular disease, cancer, diabetes, osteoarthritis, and chronic kidney disease (Ng et al., 2013). However, apart from having a disastrous impact on people’s physical health, overweight and obesity have also been linked to the increased prevalence of psychological disorders (Hill, Wyatt, Reed, & Peters, 2003). Lastly, overweight and obesity also have an excessive impact on public health care costs (Finkelstein, Ruhm, & Kosa, 2005).
the fact that supermarkets account for up to 75% of household food purchases (Glanz et al., 2012) and food purchases have proven to be strongly predictive of actual dietary intake (Ransley et al., 2003; Ransley et al., 2001).
Additionally, Hill et al. (2003) established that an intervention that leads to a decreased energy intake of just 100 kilocalories per day could cancel out weight gain in about 90% of the population. More importantly, the majority of people should be able to eat 100 kcal per day less by merely choosing a relatively healthier alternative within the same product category (Hill et al., 2003). In other words, it might not be necessary for (most) people to drastically change their eating patterns. Simply choosing, for instance, the oven-baked chips instead of the regular chips could be enough to offset weight gain. This is exactly where grocery store interventions could play an important role and eventually, could have a positive impact on dietary intake and reducing obesity (Milliron et al., 2012).
However, health interventions can also backfire and result in the purchasing of more calories instead of less (Cleeren et al., 2016; Waterlander et al., 2012). Some typical health interventions that are used nowadays, i.e. the introduction of low-fat products, price taxes, and subsidies, appear to have a negative influence on the overall healthiness of consumers’ shopping baskets (Cleeren et al., 2016; Waterlander et al., 2012). A possible explanation for this can be found in the dynamics following an intervention. For instance, when a consumer encounters a health intervention in the supermarket and decides to go for a relatively healthy product, the consumer might next license the decision for an unhealthier product.
Moreover, the long-term effects (i.e. total basket healthiness, subsequent purchasing decisions) of health interventions still remain unclear, because past research on the healthiness of food purchases has mainly been focused on single product purchases instead of sequential choices (Chandon & Wansink, 2012; Glanz et al., 2012). However, consumers in the real world often need to make decisions after they already made other choices, leading to a series of decisions where choices follow each other (Khan & Dhar, 2006). Therefore, existing knowledge in the grocery shopping domain, based on single choices or single product purchases, might be inadequate in examining possibilities to implement successful interventions and constrain calorie intake.
on shopping trip dynamics is still in its infancy. Also, despite the potential relevance of interventions, it remains unclear to date what the specific effect of an intervention on shopping trip dynamics is. Both concepts, interventions and shopping trip dynamics, have been studied separately before. However, to the best of my knowledge, no research yet has focused on the effect of an intervention on shopping trip dynamics. This gap in the literature raises some interesting questions that the current study aims to address. Primarily, if a health intervention is successful within a certain product category, can this health gain be maintained during the rest of the shopping trip? If not, might a tasty, but unhealthy intervention offer the solution and lead to the desirable health results? Answers to these questions have important practical implications for food marketers, retailers, policy makers and eventually for consumers as well.
For this reason, this paper aims to contribute to the existing literature by examining the influence of an intervention on healthy shopping dynamics. I will explain how this may be relevant. Van der Heide et al. (2016) defined healthy shopping dynamics as: “the interdependencies between the relative healthiness of sequential food choices within major shopping trips” (p.3). Additionally, relative healthiness was defined as the number of calories of a choice subtracted from the product-category average (Van der Heide et al., 2016). The current research builds on the concept of healthy shopping dynamics and contributes by examining the impact of different types of interventions, specifically the introduction of a new healthy, unhealthy (tasty) and neutral (good) cookie. More precisely, the goal of this research is to examine the subsequent choices consumers make after an intervention, thus looking at a wider range of choices instead of just two sequential choices. Therefore, the main question the current research aims to address is: how does the healthiness of the intervention choice affect the subsequent choice(s) consumers make? Also, does the type of intervention (healthy or tasty) influence the effect of the intervention?
2. THEORETICAL BACKGROUND
Choices within the food decision-making process are influenced by an abundance of factors, such as social and interpersonal influences (McFarren, 2009). Additionally, also more practical considerations such as the convenience and availability of food play a role in the food decision-making process (Swinburn et al., 2011). Moreover, health-related goals, such as a consumer’s perceived body image and the perceived health benefits of food are important drivers of food choices (Neumark-Sztainer, 1999).
In addition, as proposed by Muraven and Baumeister (2000), the food decision-making process is also defined by a need for self-control which may consume a limited resource. This resource can be depleted over time. In other words, normally, a hungry person will respond to a tasty meal by wanting to eat it since this offers immediate gratification and stills the hunger (short-term goal). But when a person is watching their weight and has to refrain him- or herself from the unhealthy, immediate and tasty meal, it requires self-control to watch your weight and consequently choose a healthier option (long-term goal). After such self-control efforts, people are more inclined to fail during the subsequent self-control attempts since the limited resource people have might be temporarily depleted (Muraven & Baumeister, 2000).
To summarize, a myriad of factors influences the healthiness of food choices that consumers make and self-control is needed in order to adhere to long-term, health-related goals. In the past, research has mainly been focused on how consumers reached a single decision or how they decided to make a single product purchase. However, many consumer choices are preceded and followed by other choices, which makes the single product purchase approach slightly unrealistic and perhaps ineffective in examining the grocery shopping process (Khan & Dhar, 2006). Nowadays, more research endorses the fact that there can be several dynamics in play within a shopping trip and therefore consumer food choices should be studied in series of choices instead of a singular choice (Gilbride et al., 2015; van der Heide et al., 2016).
These two dynamics are also possible the other way around, namely when an initial unhealthy choice is followed by either a healthy choice or another unhealthy choice.
More importantly, all four dynamics offer different explanations for consumer food purchasing behaviour. The current study aims to answer various questions regarding the effect of interventions on shopping dynamics. Does the healthiness of the intervention choice affect that of the subsequent choice? Does the type of intervention influence the healthiness of the intervention choice? Does the type of intervention perhaps also influence the effect of the healthiness of the intervention choice on that of the subsequent choice? Does the type of intervention influence the healthiness of the subsequent choice? Consequently, the different shopping dynamics will be discussed in more detail below and predictions will follow based on this.
Licensing behaviour
Goal commitment
For a long time it has been believed, especially within the social psychology domain, that consumers have a strong motivation to behave consistently and to maintain a consistent self-image (Beaman, Cole, Preston, Klentz, & Steblay, 1983). People’s need to strive for consistency can also be described as goal commitment behaviour. However, the last (nearly) two decades’ numerous studies have examined what seems to be the exact reverse phenomenon of consistency, namely licensing behaviour (Mazar & Zhong, 2010; Monin & Miller, 2001).
Although these two dynamics (licensing behaviour and goal commitment) seem to be exclusive of each other, recent studies proved there is a possibility that both dynamics exist and can alternate within a shopping trip (Mullen & Monin, 2016).
All in all, according to the goal commitment theory, it is expected that when a consumer initially chooses a healthy alternative, the consumer will subsequently choose another healthy alternative in order to maintain or achieve consistency.
Reparative behaviour
This dynamic could occur when, for instance, a consumer wants to eat healthy and tries to lose weight (long-term health goal) but instead of choosing a healthy alternative, the consumer decides to choose an unhealthy alternative. This might lead to immediate gratification, but the consumer’s long-term goal is compromised. In turn, this could lead to negative emotions and feelings of guilt, which can then trigger consumers to display reparative behaviour. This means compensating for something an individual believes he or she did ‘wrong’ by trying to repair their initial behaviour by acting less indulgent in the subsequent choices (Ramanathan & Williams, 2007). Based on the shopping dynamic of reparative behaviour, it is expected that an initial unhealthy choice, is followed by a healthier subsequent choice.
‘What-the-hell effect’
turn leads to the decision of unhealthier subsequent choices. Cochran and Tesser (1996) found evidence for the existence of this so-called ‘what-the-hell’ effect. Abandonment of the goal could also be seen as failure of self-regulation, since people give up and are not willing to put in more effort to stay committed to their goal (Mann, De Ridder, & Fujita, 2013). Hence, this dynamic leads to an initial unhealthy choice being followed by a subsequent unhealthy choice as well.
To summarize, four different dynamics could play a role within the grocery shopping process of consumers. The four possible dynamics can be found in Table 1. When a consumer initially makes a healthy choice, this may be followed by either an unhealthy choice (licensing behaviour) or another healthy choice (goal commitment). Also, when a consumer initially makes an unhealthy choice, this may be followed by either a healthy choice (reparative behaviour) or another unhealthy choice (‘what-the-hell’ effect).
TABLE 1
Possible dynamics within a shopping trip
Shopping dynamic: Initial choice Subsequent choice
Licensing behaviour Healthy Unhealthy
Goal commitment Healthy Healthy
Reparative behaviour Unhealthy Healthy
‘What-the-hell’ effect Unhealthy Unhealthy
Hence, studies regarding shopping dynamics offer evidence for predictions in both directions (healthy and unhealthy). Accordingly, the hypotheses regarding shopping dynamics are two-sided as well:
H1: Different dynamics play a role in the shopping process within a supermarket trip; H1a: An initial healthy choice is either followed by an unhealthy choice (licensing effect) or another healthy choice (goal commitment)
Additionally, to answer the question whether the healthiness of the intervention choice (i.e. the choice from the intervention category) affects the subsequent choice, two-sided hypotheses are formulated as well:
H2: The healthiness of the intervention choice affects the subsequent choice;
H2a: A relatively unhealthy intervention choice is followed by either a stronger healthy choice (reparative behaviour) or a stronger unhealthy choice (‘what-the-hell’ effect), relative to the no intervention condition
H2b: A relatively healthy intervention choice is followed by either a stronger unhealthy choice (licensing effect) or a stronger healthy choice (goal commitment), relative to the no intervention condition
Impact of interventions
consumers begin to purchase low-fat items, but also still purchase (at least) the same amount of regular items. This example illustrates the importance of studying the dynamics following an intervention. An intervention might look successful when only studying the intervention choice and the subsequent choice, but it could still attain undesirable results over the long run. Therefore, it is important to research health interventions over a longer period of time, including various consecutive purchasing decisions.
On the contrary, if a healthy intervention can backfire and lead to unhealthier subsequent choices, the reverse might also be possible. Could a tasty intervention (in this case, the introduction of an unhealthy cookie) stimulate consumers to purchase relatively healthier products afterwards? If a tasty intervention leads to more relatively healthy subsequent choices than a healthy intervention, retailers might consider revising their interventions. Consequently, knowledge regarding the intervention effect could have important practical consequences for grocery stores. Hence, in order to invent successful interventions, it could also be interesting to examine the effects of a tasty, but unhealthy intervention. Although this seems contradicting at first, it might lead to an overall healthier shopping basket at the end of the shopping trip.
Type of intervention
In order to answer the research question regarding whether the type of intervention (i.e. healthy or tasty) influences the healthiness of the choice from the intervention category (i.e. intervention choice), the following hypotheses have been formulated and tested:
H3: The type of intervention (healthy or tasty) influences the healthiness of the intervention choice;
H3a: A healthy intervention is followed by either a relatively healthy or unhealthy intervention choice
H3b: A tasty intervention is followed by either a relatively healthy or unhealthy intervention choice
H4: The type of intervention influences the effect of the healthiness of the intervention choice on that of the subsequent choice;
H4a: A tasty, unhealthy intervention leads, through either an unhealthy or healthy intervention choice, to either an unhealthy or healthy subsequent choice
H4b: A healthy intervention leads, through either an unhealthy or healthy intervention choice, to either an unhealthy or healthy subsequent choice
Furthermore, the following hypotheses have been formulated in order to test whether the type of intervention directly influences the healthiness of the choice following the choice from the intervention category (i.e. subsequent choice):
H5: The type of intervention influences the healthiness of the subsequent choice;
H5a: A tasty intervention leads to either a stronger healthy or a stronger unhealthy subsequent choice, relative to the healthy and no intervention condition
H5b: A healthy intervention leads to either a stronger healthy or a stronger unhealthy subsequent choice, relative to the unhealthy and no intervention condition
Timing of the intervention
As mentioned briefly before, the food-decision making process also requires consumer’s self-control, which is derived from a limited resource that can get depleted over time (Muraven & Baumeister, 2000). A consumer’s ability to follow his or her original shopping plan may be depleted as the shopping trip progresses (Gilbride et al., 2015). Therefore, the current study addresses the following question: does the timing of the intervention (early or late) influence the effect of the intervention?
(Baumeister, Bratslavsky, Muraven, & Tice, 1998; Muraven & Baumeister, 2000), the following hypothesis is derived:
H6: The timing of the intervention influences the effect of the intervention; an early intervention will lead to healthier subsequent choices than a late intervention
Conceptual model
Based on the literature discussed above, predictions regarding the current research questions have been formulated. Also, the conceptual framework in Figure 1 graphically displays all researched relations. The discussed predictions and displayed relations have been empirically tested and will be discussed in the results section.
FIGURE 1
Conceptual model of the current research
3. RESEARCH METHODS Design and research type
be the introduction of a new cookie. Also, the timing of the intervention is considered as an experimental factor in the research design. Overall, there are four intervention levels (control, neutral, healthy, and unhealthy) and two timing levels (early and late introduction). This results in a total of eight conditions, as can be seen in Table 2. Furthermore, the chosen study design is a between-subjects design which means that each participant is tested under one condition only. As a result, each group consisted of around 36 participants.
TABLE 2
Eight conditions constituting of combinations of two experimental factors: timing of the intervention and type of intervention
Timing of the intervention:
Control Unhealthy cookie introduction Neutral cookie introduction Healthy cookie introduction Early (3rd category) Control (no intervention) Early introduction of unhealthy cookie Early introduction of neutral cookie Early introduction of healthy cookie Late (11th category) Control (no intervention) Late introduction of unhealthy cookie Late introduction of neutral cookie Late introduction of healthy cookie Participants
A random sample of participants was recruited and asked to participate via Amazon Mechanical Turk. A total of 326 people started the online questionnaire of which 289 were eventually included in the research. Two exclusion criteria were maintained. If participants either did not live in the USA or were not the primary grocery shopper in their household, they were immediately skipped to the end of the survey and excluded from the research.
Furthermore, the average age of the participants, of which 69% were female and 31% were male, vary between the age of 18 and 76 (M = 42,59, SD = 13,21). In addition, the average household size consisted of 3 persons. Lastly, the median of monthly income per household was 3500 dollars. However, it must be noted that 63 participants decided not to answer this question. Procedure
surrounding the intervention, whereas Vos’s research is more focused on the effect of the intervention on the healthiness of the shopping basket at the end of the shopping trip. Additionally, both studies also used different dependent variables and different statistical analyses in order to test the hypotheses. Moreover, the current study is also focused on modelling the food choices and examining the effects by means of panel regressions, whereas the research of Vos relied solely on a cross-sectional analysis by means of a three-way ANOVA.
The data that is used for the analyses in this research is from an online questionnaire that included the simulation of a grocery shopping environment and a survey. The questionnaire started with a brief introduction describing what the participants could expect in the survey. The introduction also contained the estimated time needed to complete the survey (approximately 10 minutes), information regarding the prize package, and a promise that all data is kept strictly confidential and is processed anonymously.
After the introduction, the two exclusion criteria questions were asked. Subsequently, all participants who were not excluded from the research were randomly assigned to one of the eight conditions. Six of the conditions contained a type of intervention, either healthy, unhealthy or neutral, and two conditions served as control conditions.
Subsequently, all participants were asked to shop for 16 food products within 16 different product categories. Within each product category, three different options were displayed. Except for the intervention conditions in which case the introduction of a new cookie presented a fourth option. The participants were asked to choose only one option per category and to choose the option that they would most likely purchase when in a real store.
Following the shopping task, participants within six out of the eight conditions (all except the two control conditions) received several follow-up questions regarding the introduction of the new cookie. For instance, ‘What did the banner in the picture of the new option read?’. These questions were included to assess whether participants actually read the banner and whether they consciously processed the new information.
Stimuli
The questionnaire used to gather data was an online questionnaire which was distributed via Amazon Mechanical Turk. The complete questionnaire can be found in Appendix A. The questionnaire started with a brief introduction, followed by the shopping task (one of the eight conditions), then possibly a few follow-up questions regarding the banner, and lastly, several questions regarding participants’ demographics and information about their health and grocery shopping patterns and experiences.
Prize package. As mentioned briefly before, a prize package worth of $75 in value was included to align the incentives of participants and to motivate participants to fill in the questionnaire as realistic as possible (Ding, 2007; Van Ittersum, Wansink, Pennings, & Sheehan, 2013). All participants had an equal chance of winning the prize package. In order to motivate participants to purchase the options they would most likely purchase on a real shopping trip, the prize package consisted out of the items in your shopping basket and the remaining cash from the $75. In other words, if you win the prize package and your final basket is relatively cheap, you will receive more money in cash than if you would have chosen more expensive products during the shopping task. Food products. The decision for the sixteen product categories was primarily based on the United States Department of Agriculture (USDA) Thrifty Meal Plan (USDA, 2000). The Thrifty Meal Plan describes how a diet that meets the minimum recommendations of the 1995 Dietary Guidelines for Americans may be accomplished by a family consisting of four people that have a modest budget or receive food stamps (Jetter & Cassady, 2006). Some product categories were not derived from the Thrifty Meal Plan but were nonetheless included in the questionnaire since they represent categories from which people usually shop.
Additionally, within each product category three different alternatives were displayed. Namely, one healthy option, one unhealthy option, and one neutral option. The decision for a neutral option has been made to expand the possible alternatives and to offer a middle way as well. Furthermore, the three alternatives within each product category are differentiated by means of the classification of the Traffic Light Labeling, which has been created by the Food Standards Agency (2015).
brand. However, within four categories this was not possible because the brand did not offer a relatively healthy or relatively unhealthy alternative. Additionally, in order to avoid selection based on taste preference, almost all products within a category had the same ‘flavour’. For instance, within the category salad dressing, all three alternatives carried a ‘ranch’ flavoured salad dressing. The sixteen product categories including the three alternatives per product category are presented in Table 3 below.
TABLE 3
Sixteen product categories with corresponding product alternatives
Product category:
Unhealthy product Neutral product Healthy product
Beef patties Ground Beef Patties 73% Lean Ground Beef Patties 85% Lean Ground Beef Patties 93% Lean Breads Great Value White Bread Great Value Brown Bread Great Value Whole Wheat
Bread Breakfast
Biscuits
belVita Chocolate Biscuits belVita Golden Oat Biscuits belVita Mixed Fruits Biscuits
Chips Chips Pringles Original Potato Crisps
Pringles Original Reduced Fat
Potato Crisps Pringles Original Fat Free Potato Crisps
Cola Pepsi Cola Pepsi Next Pepsi Diet
Cold Cereal Great Value Sugar Frosted
Flakes Great Value Corn Flakes Great Value Bran Flakes Cookies Chips Ahoy! Chunky
Chocolate Chip Cookie
Chips Ahoy! Original Chocolate Chip Cookie
Chips Ahoy! Oatmeal Chocolate Chip Cookie Ice Cream Ben & Jerry’s Chocolate Ice
Cream
Ben & Jerry's Greek Frozen Yoghurt
Ben & Jerry's Frozen Yoghurt Half Baked
Macaroni & Cheese
Hospitality Mac & Cheese Dinner
Kraft Premium Three Cheese Macaroni & Cheese Dinner
Annie’s Homegrown Organic Classic Mild Cheddar Macaroni & Cheese
Milk Great Value Fat 2% Milk Great Value Low Fat 1% Milk Great Value 0% Fat Milk Pasta Sauce Newman's Own Pasta Sauce Bertolli Organic Traditional
Sauce Gina Rispoli All Natural Sauce Peanut Butter Skippy Creamy Peanut Butter Skippy Natural Creamy Peanut
Butter Skippy Reduced Fat Creamy Peanut Butter Salad
Dressing
Great Value Creamy Ranch Salad Dressing
Great Value Classic Ranch Light Dressing
Great Value Fat Free Creamy Ranch Salad Dressing Spaghetti Great Value Spaghetti Great Value Whole Wheat
Spaghetti Barilla Whole Grain Spaghetti Spreadable
Butter
Land O Lakes Butter with Canola oil
Land O Lakes Butter with Olive Oil
Land O Lakes Butter Light
Yoghurt Mountain High Original Plain Style yoghurt
Mountain High Plain Low-fat Yoghurt
Mountain High Plain Low-fat Yoghurt
all three alternatives (i.e. relatively healthy, relatively unhealthy and neutral). Within the control conditions only the three regular cookie alternatives were included. However, in the experimental conditions participants were offered a fourth cookie option, namely a new cookie. Within the neutral, unhealthy, and healthy conditions the cookie banners respectively read ‘New & Good’, ‘New & Tasty’, and ‘New & Healthy’. An overview of the three new cookies and their banners can be found in Appendix B.
Timing. Furthermore, in order to test the influence of the timing of the intervention on the effect of the intervention, two experimental conditions were created. Half of the participants received the cookie intervention as the 3rd product category choice (early condition) and half of the participants
received the intervention as the 11th product category choice (late condition). All other categories were presented to the participants in randomized order to avoid bias.
Psychographics. After the shopping task, participants had to choose whether they strongly disagreed or strongly agreed with several statements. The participants had to rate these statements based on a seven-point Likert scale (Likert, 1932), which ranged from 1 (strongly disagree) to 7 (strongly agree). The fourth option presented a neutral answer option namely ‘neither agree nor disagree’. These statements were included to measure the health consciousness of participants and to gain more insights in the shopping patterns and experiences of the participants. Health consciousness was measured based on six items that were derived from a study by Dutta-Bergman (2004). The following items were used: ‘Living life in the best possible health is important to me’, ‘Eating right, exercising, and taking preventive measures will keep me healthy for life’, ‘My health depends on how well I take care of myself’, ‘I actively try to prevent disease and illness’, ‘I do everything I can to stay healthy’, and ‘Caloric information is important to me’.
Measurement of variables
In the current study, the sequence of food decisions by each participant is observed, and the goal is to determine the influence of an intervention on the relative healthiness of subsequent choices.
(per 100 gram) of a product choice divided by the product-category average. However, in the current research relative healthiness has been operationalized differently. Relative healthiness has now been defined as the number of calories (per 100 gram) of a product choice subtracted from the product-category average. For instance, the product category chips contains three different alternatives, namely ‘Pringles original potato crisps’, ‘Pringles original reduced fat potato crisps’, and ‘Pringles original fat free potato crisps’. These alternatives respectively contain 516 calories, 499 calories and 250 calories per 100 grams. This results in a product category average of 421 calories. If a consumer chooses the fat free alternative, the relative healthiness will be calculated as follows: 421 - 250 = 171. All values in the original dataset have been recoded and transformed to the new operationalization of the dependent variable relative healthiness. In conclusion, higher positive values of relative healthiness depict healthier choices in comparison to the product category average.
Lagged relative healthiness. Since the goal of this study is to examine a sequence of decisions, the interdependencies between these choices need to be taken into account by means of a lagged variable for relative healthiness. The lags of relative healthiness (n-1) depicts the previous choice and will be included in the analysis as an independent variable. Notably, a difference was created between relatively healthy and relatively unhealthy previous choices. Specifically, the relative healthiness of the previous choice was split into two variables, choices above 0 (i.e. positive values which means relatively healthy choices) and below 0 (i.e. negative values which means relatively unhealthy choices). The lags of the relatively unhealthy previous choices have been recoded for interpretation, such that these values are not negative anymore. Consequently, higher positive values for a relatively unhealthy previous choice indicate unhealthier previous choices relative to the product category average.
Control variables. The current model will control for several variables in order to examine whether the intervention effect is actually due to the intervention and not to other factors that possibly influence this effect. A product characteristic for which will be controlled is price. The original dataset did not contain this variable yet. Price index is calculated by comparing the relative price of a choice to the average price of all products in a specific product category. Furthermore, the demographic variables for which will be controlled are age, gender, weight, and monthly income. Notably, the original dataset included both the variables height and weight. However, in the current research BMI is calculated and used for further analysis. Lastly, the model will also control for the individual health consciousness scores of participants. More specifically, the mean-centered sum of all six items, derived from the study of Dutta-Bergman (2004), will be computed and used in the analysis. The variable is mean-centered because the variable has been measured on a Likert scale (ranging from 1 till 7). Consequently, 0 would have no meaning in the regression analyses. In conclusion, an overview of the variables discussed above and their operationalization can be found in Table 4.
TABLE 4
Variable operationalization panel regressions
Variable: Description:
RHin Relative healthiness of product choice n in shopping basket i, calculated by subtracting the calorie count of choice n from the category average calorie count (dependent variable in most analyses)
RHhi,n-1 Relative healthiness of relatively healthy (> 0) previous choice n-1 in shopping basket i
RHui,n-1 Relative healthiness of relatively unhealthy(< 0) previous choice n-1 in shopping basket i; reverse-coded such that more positive values indicate unhealthier choices
Cookie Cookie dummy indicating 1 for the intervention category choice and 0 for all other choices; hence per individual only one 1 and fifteen 0’s
Cookie n-1 Lagged cookie dummy indicating 1 for the choice after the intervention category choice (4th or 12th choice)
Cookie n-2 Two lags cookie dummy indicating 1 for the second choice after the intervention category choice (5th or 13th choice)
Cookie_both Variable is a combination of Cookie and Cookie n-1; indicating 1 for the intervention and subsequent choice and 0 for all other choices (remaining 14) Healthy Intervention dummy indicating 1 for the intervention category choice within
Tasty Intervention dummy indicating 1 for the intervention category choice within the tasty intervention condition and 0 for all other choices
Control variables
PIin Price index of choice n in shopping basket i, measured as the price of choice n divided by the category average price
HealthConsi Health consciousness score per individual based on the sum of the six items that measure this construct; variable is mean-centered
BMIi BMI indicates the BMI per individual and is calculated by the formula: weight (lbs) / [height (in)]2 x 703
Plan of analysis
The required sample size for this study was determined through the program G* Power. According to the power analysis of G* power, 270 participants were needed in order for this study to have sufficient power (see Appendix C). Consequently, a sample of 289 participants was estimated to ensure (more than) 80% power at a 0.05 significance level (2-sided) to detect an effect of d = 0.25.
Statistical analyses in the current study were performed using three programs because the various types of analysis required different programs and packages. More specifically, the programs R Studio (version 1.1.383; R Core Team, 2017), SPSS (version 25), and Stata (version 15) have been used. Also, the significance level that has been maintained for all analyses is 0.05. However, before any analysis could be conducted, the data needs to be cleaned and the variables need to be properly coded. Preparing the data also entails analysing the missing values and checking for outliers. Another step in the data preparation process was to also generate a dataset which was in long format. The original dataset was in wide format, which is required for analyses of variances (ANOVA). However, in order to conduct panel regressions, the data needed to be in long format.
Furthermore, the nature of the data that is analysed is panel data, which is also called longitudinal or cross-sectional time-series data (Wooldridge, 2010). Accordingly, the data will be analysed by means of hybrid panel regressions which allow for fixed and random effects within the data (Allison, 2009). The choice for a hybrid model has been made because random-effects allow for time invariant variables (i.e. gender) and a fixed-effects model fits better with lagged variables (Wooldridge, 2010).
Lastly, several analyses of variance (ANOVA) will be conducted in order to test the effects of the type and timing of the intervention on the relative healthiness of the intervention choice and subsequent choice. However, before these analyses can be conducted, several assumptions need to be tested. Also, it is important to note that the variable operationalization is slightly different for the ANOVA’s than for the panel regressions. An overview of the variable operationalization can be found in Table 5.
TABLE 5
Variable operationalization ANOVA
Variable Description
RH_CookieChoice Relative healthiness of the cookie choice (3rd or 11th); calculated by
subtracting the calorie count of the cookie choice from the cookies category average calorie count
RH_SubsequentChoice Relative healthiness of the choice after the intervention choice (4th or the 12th); indicating the relative healthiness of the subsequent choice
InterventionType Type of intervention indicating which intervention an individual received: control condition (0), unhealthy cookie (1), neutral cookie (2), and healthy cookie (3)
Timing Timing dummy indicating whether the introduction of the new cookie was early (0) or late (1)
The current methodology also carries some potential limitations, which will be discussed further in the discussion section.
4. RESULTS
individual. The modified dataset also includes several demographics (i.e. age, gender, monthly income, BMI), a price index variable and an individual (mean-centered) health consciousness score. The structure of this chapter will be as following. First, a reliability analysis regarding the health consciousness construct and several descriptive statistics will be given. Next, an overview of the correlations between the most important variables in this study will be discussed, followed by panel regressions and several ANOVA’s in order to test the specific hypotheses. Lastly, an overview of the hypotheses will be provided indicating which hypotheses are rejected and which ones can be accepted.
Reliability analysis health consciousness
The construct ‘health consciousness’ is measured by means of six items. A reliability analysis needs to be conducted in order to check if the questions capture the construct well and if internal consistency can be guaranteed. The full results of this reliability analysis can be found in Appendix D and an overview of the results in Table 6. Usually, a Cronbach alpha of 0.7 and above is considered to be a good indicator of reliability (Santos, 1999).The reliability analysis showed that the overall Cronbach’s alpha of health consciousness is 0.86, which means that the construct is highly reliable. In addition, this analysis showed that dropping an item does not increase the overall alpha of the scale, which is also positive.
TABLE 6
Reliability analysis health consciousness
Items Alpha Alpha if item is dropped N Mean SD
Scale 0.86 n/a 289 5.6 0.98 Health_1 0.84 0.81 289 5.8 1.2 Health_2 0.74 0.84 289 5.8 1.1 Health_3 0.65 0.85 289 6.0 1.0 Health_4 0.82 0.82 289 5.6 1.3 Health_5 0.80 0.83 289 5.1 1.4 Health_6 0.76 0.85 289 5.1 1.6 Descriptive statistics
TABLE 7
Descriptive statistics sociographic variables
Variables Mean SD Min Max
Age 42.47 13.40 18 76 Gender 1.69 0.46 1 2 BMI 27.47 7.67 10.18 59.51 Monthly income 17111.21 29876.68 0 250000 Household size 3.06 1.48 1 8 Health consciousness 5.57 0.98 2 7
Additionally, to gain some primary insights in the distribution of the dependent variable relative healthiness, a graph was made containing the means of all eights conditions (see Figure 2). First of all, it is noticeable that all group means are negative which indicates that, on average, participants made more relatively unhealthy food choices compared to the product category average. Also, it appears that participants in the early timing and unhealthy intervention condition had the highest, negative mean regarding the relative healthiness of all 16 food choices. Additionally, participants in the late, control condition obtained the lowest mean, which means that, on average, they had the healthiest shopping baskets at the end of the shopping trip.
FIGURE 2
Mean of the relative healthiness of the complete shopping trip per condition
Correlations
Also, in order to gain more insights in the variables and as a first step in performing a regression analysis, correlations between the relevant variables are checked. An overview of these correlations together with the means and standard deviations of relative healthiness (RH), relatively healthy previous choice (RHhi,n-1) and relatively unhealthy previous choice (RH
u
i,n-1) can
be found in Table 8. All three possible correlations are found to be significant, which indicates that the variables are associated with each other. More specifically, the variables RH and RHui,n-1 are
significantly correlated, r = -.10, p < .01. Also, RHhi,n-1 and RH u
i,n-1are correlated, r = -.28, p < .01.
Lastly, the correlation between RH and RHhi,n-1 is also found to be significant, r = .13, p < .01.
TABLE 8
Descriptive statistics and correlations
Variables Mean SD 1 2 3
1 RHin 4.51 95.97
2 RHhi,n-1 34.38 51.85 0.13***
3 RHui,n-1 29.58 67.12 -0.10*** -0.28***
Hybrid panel data
After exploration of the data and several initial analyses, the first hypotheses can be tested by means of panel regressions. In order to account for the panel data structure in the dataset, the regression analyses will be conducted with a hybrid panel data model (Allison, 2009), which is originally derived from Mundlak (1978). The hybrid panel model combines the advantages of a random-effects model and a fixed-effects model (Schunck, 2013). More specifically, a hybrid panel data model is able to estimate a model with both varying (i.e. price index) and time-stable variables (i.e. gender). Although random effects regression carries the assumption that the observed variables are uncorrelated with the unobserved variables, Mundlak (1978) proposed a technique in order to relax this assumption.
Additionally, the within transformation of variables contains the values in deviation from the individual means, which can also be called group mean-centering (Allison, 2009).
Another important characteristic of this model is the inclusion of an autocorrelation coefficient (Φ), which was also done in the study of Gilbride, Inman, and Stilly (2015). The autocorrelation coefficient ensures that the estimates will not be biased if the error terms are serially correlated due to other factors than the included independent variables.
Model building
Several panel regressions have been executed in order to test the hypotheses. Relative healthiness (RH) was the dependent variable in all cases. The initial model also controlled for age, gender, income per month and household size. However, these predictors did not have a significant impact on the outcome variable and in addition, were contextually less relevant for the analysis. Therefore, the decision was made to exclude these demographics from the model and continue with a more parsimonious model.
Moreover, the most basic model, which has been gradually extended, consisted out of the following independent variables: health consciousness (mean-centered), BMI and the transformed within and between variables for price index, relatively unhealthy previous choice and relatively healthy previous choice.
Model estimation
TABLE 9
Regression results basic model
(1) Basic Model Parameter SE Within-effects β1 RHhi,n-1 - RHhi -0.0525*** (0.0199) β2 RHui,n-1 - RHui 0.0930*** (0.0206) Between-effects β3 RHhi 0.938*** (0.113) β4 RHui -0.863*** (0.138) Control variables β5 HealthConsi -0.427 (0.866) β6 PIin 39.72** (19.71) β7 PIin - PIi 179.7*** (4.393) β8 BMIi -0.0352 (0.108) β0Constant -39.86* (20.84) Φ autocorrelation coefficient 0.071 Observations 4,335 Number of PPN_ID 289 *** p<0.01, ** p<0.05, * p<0.01
Intervention dummy model. The initial, basic model has been extended in order to test the effect of the intervention on the shopping dynamics of consumers. This model included a dummy which indicated 1 at the time of the intervention choice (either the 3rd or the 11th choice; depending on the timing condition) and 0 at all other choices. Moreover, interaction terms of the dummy in combination with a relatively (un)healthy previous choice were included in the model. However, as can be seen in Table 10, the results of this analysis showed that the intervention did not have a significant effect on the relative healthiness of food choices (p = .732). Additionally, the interaction terms for a relatively healthy previous choice (p = .756) and a relatively unhealthy previous choice (p = .543) were also found to be insignificant.
(p = .287) and there were no significant interaction effects (p = .913 for healthy and p = .304 for unhealthy).
Two lags intervention dummy model. Additionally, a model was estimated with a dummy for the second choice participants made after the intervention choice. In other words, this dummy represented the 5th or 13th choice consumers made in the online simulated grocery store, depending on the condition they were in. Similarly, these results also indicated there was no significant (lagged) intervention effect (p = .089). However, as can be seen in Table 10, the (lagged) intervention effect would have been significant at an p < 0.1 significance level. In that case, the estimate of this parameter should be interpreted as following: the introduction of a new cookie leads to a healthier second choice after the choice made in the intervention category (n-2), β = 7.285, p = .089. Still, there were no interaction effects for a relatively healthy previous choice (p = .258) and a relatively unhealthy previous choice (p = .154).
TABLE 10
Regression results intervention dummy
(1) Intervention dummy model (2) Lagged intervention dummy model (n-1) (3) Two lags intervention dummy model (n-2)
Parameter SE Parameter SE Parameter SE
β7 PIin 39.78 ** (19.73) 41.15 * (21.01) 45.05 ** (21.59) β8 BMIi -0.0352 (0.108) -0.0628 (0.115) -0.0799 (0.118) β0 Constant -39.87 * (20.86) -40.19 * (22.21) -43.15 * (22.84) Φ autocorrelation coefficient 0.071 0.093 0.089 Observations 4,335 4,046 3,757 Number of individuals 289 289 289
Cookie* intervention effects (within both the main and interaction effects): in model 1 Cookie (intervention choice), in model 2 Cookien-1 (lagged intervention choice) and in model 3 Cookien-2 (two lags intervention choice).
*** p < 0.01, ** p < 0.05, * p < 0.1
Intervention and lagged intervention dummy model. This model included a dummy which indicated a 1 at the time of the intervention and subsequent choice and 0 at all other choices. Hence, per individual this dummy contained two 1’s and fourteen 0’s. However, as can be seen in Table 11, this analysis also indicated there was no (lagged) intervention effect (p = .720). Also, the interaction effect of a relatively healthy previous choice (p = .244) and a relatively unhealthy previous choice (p = .650) were also insignificant.
TABLE 11
Regression results subsequent and intervention dummy, healthy dummy and tasty dummy
(1) Intervention choice and subsequent choice (both) dummy model
(2) Healthy intervention
dummy model
(3) Tasty intervention dummy model
Parameter SE Parameter SE Parameter SE
As a final point, it is not repeated in the discussion of each separate panel regression analysis, however the effects of PI, health consciousness and BMI remain relatively consistent across all analyses. In conclusion, PI has a significant impact on relative healthiness in all analyses, meaning that a higher price index relates to healthier product choices. Additionally, health consciousness and BMI do not significantly affect the relative healthiness of food choices within this study. Lastly, the autocorrelation coefficient is also quite stable across the different analyses. The autocorrelation coefficient varies between Φ = 0.071 and Φ = 0.093, which indicates that the error terms are represented by a relatively low, positive autocorrelation.
Robustness check
In order to see whether the models estimate consistent results, a robustness check, which was similar to the robustness check in the paper of van der Heide et al. (2016), was performed. Namely using the single lag variable for relative healthiness (RHi,n-1) instead of the two separated
lag variables (RHhi,n-1 and RHui,n-1) that have been used in the other models. The model with the
single lag variable yields similar results as the models estimated with the two lag variables, as can be seen in Table 12. More specifically, the healthiness of the previous choice (RHi,n-1) is negatively
associated with the dependent variable that depicts the relative healthiness of the current choice (β = -.072, p <001). Since these findings are in line with previous findings (using the two lag variables), the robustness check shows that the model is robust and the estimates remain consistent.
TABLE 12
Robustness check single lag variable
Additionally, a second robustness check was conducted to see whether the findings still remain consistent under different conditions. The hybrid panel data model with ‘mean’ and ‘diff’ variables for the time-variant variables, should yield approximately the same results as the original variables (without mean and diff transformation) in a fixed effects regression. As can be seen in Table 13, there appeared to be slight differences in the parameters and significance levels of the ‘diff’ variables in the hybrid panel regression and the original (not-transformed) variables in the fixed effects regression. However, a logical explanation for this could be that all stable variables are omitted from the analysis (i.e. BMI and health consciousness). Concluding, since there are only small differences and these are probably largely the result of the omitted variables, the parameters can be interpreted as reliable.
TABLE 13
Robustness check fixed effects regression
(1) Fixed Effects Parameter SE β1 RHhi, n-1 -0.0417** (0.0194) β2 RHui, n-1 0.0678*** (0.0200) β3 HealthConsi 0 (0) β4 PIin 267.9*** (5.152) β5 BMIi 0 (0) β6 Cookie* -4.077 (6.489) β7 Cookie* x RHhi 0.169 (0.234) β8 Cookie* x RHui 0.228 (0.267) β0 Constant -269.3*** (4.863) Φ autocorrelation coefficient 0.051 Observations 4,046 Number of PPN_ID 289 *** p<0.01, ** p<0.05, * p<0.1 ANOVA
a between-subjects design is used since the subjects are assigned to only one of the eight conditions.
Before conducting the analyses, several assumptions need to be tested in order to check if an ANOVA would produce reliable results. The first requirement that needs to be met is that there should not be any significant outliers, since this can reduce the accuracy of the results. This assumption has been checked by means of box plots, which showed there are no significant outliers.
Secondly, the dependent variable needs to be approximately normally distributed for each combination of the groups of the independent variables. The Shapiro-Wilk and Kolmogorov-Smirnov test of normality were both significant (p < .05), which means that the null hypothesis is rejected and the normality assumption is violated (see Appendix E). Furthermore, a visual inspection of Normal Q-Q plots confirmed this and also showed that both dependent variables (intervention choice and subsequent choice) are not normally distributed (see Appendix F). ANOVA tolerates violations to the normality assumption rather well, however, it should be kept in mind that the results of the ANOVA can be slightly inaccurate.
Finally, the last assumption that has been tested was the assumption of homogeneity of variance. This has been tested by means of the Levene’s test for homogeneity of variances (p = .233), which showed that the variance of the dependent variable ‘relative healthiness of the subsequent choice’ is equal across the groups. However, the Levene’s test for homogeneity of variances (p = .021) of the dependent variable ‘relative healthiness of the cookie choice’ was found to be significant, which means that the assumption of equal variance is violated. In order to account for this violation, two separate One-Way ANOVA analyses have been conducted. Cookie choice in combination with timing resulted in an insignificant Levene’s statistic (p = .142), which means that the significance level of an ordinary ANOVA is reliable. However, the Levene’s statistic of the One-Way ANOVA between cookie choice and intervention type appeared to be significant (p = .021). In order to account for this violation, a Welch ANOVA is conducted which presents a more reliable significance level in case of violation (Moder, 2010). The Levene’s tests and Welch ANOVA can be found in Appendix G.
means (regarding relative healthiness of the intervention choice) of individuals within the four different intervention conditions, F (3, 285) = .488, p = .691. Similarly, there are also no significant differences between the two timing conditions, F (1, 285) = .196, p = .658. The results of this ANOVA are presented in Table 14.
TABLE 14
ANOVA with the relative healthiness of the cookie choice as dependent variable
Variables Mean square dF F Sig. Partial Eta
Squared
Intervention_Type 377.072 3 0.488 0.691 0.005
Timing 151.311 1 0.196 0.658 0.001
Subsequently, the intervention effect is also measured by the relative healthiness of the choice after the intervention choice, also called the subsequent choice. As can be seen in Table 15, these results also showed there are no differences in group means between the different types of interventions, F (3, 274) = .353, p = .787. Furthermore, the results also showed there is no significant interaction effect between the healthiness of the intervention choice and the type of intervention, which means that the healthiness of the intervention choice does not have a moderating role on the relationship between the type of intervention and the subsequent choice, F (1, 274) = 1.085, p = .299.
TABLE 15
ANOVA with the relative healthiness of the subsequent choice as dependent variable
Variables Mean square dF F Sig. Partial Eta
Squared Intervention_Type 1609.08 3 0.353 0.787 0.004 Healthiness_Cookie Choice 6355.042 10 1.396 0.182 0.048 Healthiness_Cookie Choice * Intervention_Type 4937.75 1 1.085 0.299 0.004
R squared = 0.054 (Adjusted R squared = 0.005)
Consumers in the control condition (no intervention) choose, on average, a relatively healthier subsequent product. Also, the means of the relative healthiness of the cookie choice show that consumers in the control condition, on average, chose the most least healthy.
FIGURE 3
Relative healthiness of the cookie choice and the subsequent choice per intervention type
Hypotheses
After analysing and discussing all the results, the hypotheses are reviewed to see which hypotheses can be accepted and which should be rejected. As can be derived from Table 16, the first hypothesis, regarding the different shopping dynamics within a supermarket trip, can be supported by means of the panel regressions presented in Table 9. However, hypotheses H1a and H1b can only be partly supported since the results of this study found proof for two out of the four shopping dynamics. Namely, the results of the panel regressions (Table 9, 10, 11) showed evidence for the licensing effect and reparative behaviour, but not for goal commitment and the ‘what-the-hell’ effect. Furthermore, hypotheses 2 until 6, including their sub-hypotheses, could not be supported based on the results from this study.
-6,81 -1,04 -4,21 -8,24 -1,46 -7,85 -5,53 -6,38 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 RH_CookieChoice RH_SubsequentChoice
TABLE 16
Evaluation of the six hypotheses after analysing the results
Hypotheses: Outcome:
H1: Different shopping dynamics play a role in the shopping process within a supermarket H1a: An initial healthy choice is either followed by an unhealthy choice (licensing effect) or another healthy choice (goal commitment))
H1b: An initial unhealthy choice is either followed by a healthy choice (reparative behaviour) or another unhealthy choice (‘what-the-hell’ effect)
Supported Partly supported Partly supported H2: The healthiness of the choice from the intervention category (=intervention choice)
affects the subsequent choice
Not supported H2a: A relatively unhealthy intervention choice is followed by either a stronger healthy
choice or a stronger unhealthy choice, relative to the no intervention condition
- H2b: A relatively healthy intervention choice is followed by either a stronger unhealthy
choice or a stronger healthy choice, relative to the no intervention condition
-
H3: The type of intervention (healthy or tasty) influences the healthiness of the intervention choice
Not supported H3a: A healthy intervention is followed by either a relatively healthy or unhealthy choice
from the intervention category
- H3b: A tasty intervention is followed by either a relatively healthy or unhealthy choice from
the intervention category
-
H4: The type of intervention influences the effect of the healthiness of the intervention choice on that of the subsequent choice
Not supported H4a: A tasty, unhealthy intervention leads, through either an unhealthy or healthy
intervention choice, to either an unhealthy or healthy subsequent choice
- H4b: A healthy intervention leads, through either an unhealthy or healthy intervention
choice, to either an unhealthy or healthy subsequent choice
-
H5: The type of intervention influences the healthiness of the choice following the intervention choice (i.e. the ‘subsequent’ choice)
Not supported H5a: A tasty intervention leads to either a stronger healthy or a stronger unhealthy
subsequent choice, relative to the healthy and no intervention condition
- H5b: A healthy intervention leads to either a stronger healthy or a stronger unhealthy
subsequent choice, relative to the unhealthy and no intervention condition
-
H6: The timing of the intervention influences the effect of the intervention; an early intervention will lead to healthier subsequent choices than a late intervention
