A Ride on the Consumer’s Rollercoaster of Choices:
Predicting Healthy Shopping Behavior
2
Master Thesis
A Ride on the Consumer’s Rollercoaster of Choices:
Predicting Healthy Shopping Behavior
Date
14
thof June
Name
L.M. Grondsma
Student number
2228947
Educational program
Master in Marketing Intelligence and Marketing Management
Department
Department of Marketing
Educational institution
Rijksuniversiteit Groningen
Address
Peizerweg 18A,
9726 JJ Groningen
Phone number
06 37440748
Email address
lindagrondsma@hotmail.com
1
stSupervisor
Prof. Dr. ir. K. van Ittersum
2
ndSupervisor
Prof. Dr. T.H.A. Bijmolt
Company
Plus Retail
3
EXECUTIVE SUMMARY
As a result of the problem of obesity, a trend towards the promotion of healthier choices has evolved over the past years. This problem is largely driven by overconsumption of unhealthy foods. A good starting point to solve the problem is where people purchase their food products: grocery stores. Many studies have been performed there on product-level, but due to recent technological developments it now possible to take the entire shopping trip into account. Preliminary studies have shown that there is reason to believe that the healthiness of choices across the shopping trip evolve, which are called ‘Healthy Shopping Dynamics (HSD)’. The purpose of this research is to discover how such dynamics evolve, what influences healthy shopping baskets and whether such dynamics can be used to forecast the healthiness of future choices. This gives many insight in customers’ healthy shopping behaviors.
To perform this research, basket-level scanner data was made available by Plus, one of the largest grocery retailers in the Netherlands. This data was used to uncover drivers of healthy shopping baskets and to find out what determines the healthiness of the next purchase could be performed.
The results of this study suggest that several drivers of the healthiness of shopping baskets and of the healthiness of the next purchase can be distinguished. In particular the drivers of HSD seem to have an impact on the healthiness levels. No effects were found for the three other drivers: general promotions, health labels and economic health interventions. Most importantly it seems that healthy shopping dynamics evolve first positively towards healthy behavior after the first few choices. However, after a number of choices there is a tipping point and behavior becomes oriented towards unhealthier choices, perhaps due to licensing effects where customers allow themselves to make unhealthier choices if they already made healthy ones before. Moreover, the insignificance of general promotions and health labels contradicts a great deal of existing literature.
All these findings have some specific implications for Plus, but also broader implications for the entire grocery retailing sector. This study gives new insights in how healthy shopping dynamics evolve over the course of a shopping trip and that analysing this type of data has a lot of potential for future research.
4
PREFACE
“What is success? I think it is a mixture of having a flair for the thing that you are doing; knowing that is not enough, that you have got to have hard work and a certain sense of purpose”
Margaret Thatcher
When I started studying at the University of Groningen in 2011, I was not sure what degree would fit my interests. I choose to follow the bachelor Business Economics, which is where I first got in touch with marketing. Soon my interest for marketing began to grow and in February 2015 I started the masters Marketing Management and Marketing Intelligence. Together, these tracks have taught me several facets of the marketing field. I developed a passion for data analytics and am truly grateful that I received the opportunity to put this passion into practice with this thesis. For this, I owe much to my second supervisor prof. dr. T.H.A. Bijmolt, who suggested this project to me in the first place. I thank both him and my first supervisor prof. dr. ir. K. van Ittersum for their thorough guidance, feedback and their pure interest for the research I performed. Moreover, I would like to thank Plus for the opportunity to use their data and to thank Astrid Westerveld Msc. and Marco Maatman Msc. for their support, feedback and interesting insights. In addition to them, I also would like to thank my research partner Rutmer Faber. Even though we wrote two different theses, I appreciate the time we spent together to help each other when needed.
This is also a good opportunity for me to thank my parents, Klaas Jan and Sita, and my sister Daniëlle for their ongoing support during my study time in Groningen. Thank you for always believing in my capabilities and giving me the chance to develop myself into the person I have become. My friends have been of great value too and I would like to thank my best friend Hetty in particular. Not only for her help with the development of this thesis, but also for making my study time here in Groningen unforgettable. Finally, I am grateful for all the support I have received from my boyfriend Ricardo. Whenever I was stressed, tired or just very enthusiastic, you were always by my side.
To you who is holding this thesis now: I hope you enjoy reading it!
5
TABLE OF CONTENT
EXECUTIVE SUMMARY ... 3 PREFACE ... 4 1. INTRODUCTION ... 7 1.1 RESEARCH QUESTIONS ... 8 1.2 RELEVANCE ... 10 1.3 OUTLINE ... 11 2. THEORETICAL FRAMEWORK ... 12 2.1 CONCEPTUAL FRAMEWORK ... 122.2 HEALTHY SHOPPING DYNAMICS... 13
2.2.1 HEALTH INDEX OF THE FIRST PURCHASE DECISION ... 14
2.2.2 HEALTH INDEX OF THE PREVIOUS PURCHASE DECISION ... 14
2.2.3 TREND ... 15
2.2.4 PEAKS ... 16
2.2.5 VOLATILITY ... 16
2.2.6 AVERAGE HEALTHINESS OF THE BASKET SO FAR ... 17
2.3 DRIVERS OF HSD ... 18
2.3.1 GENERAL PROMOTIONS ... 18
2.3.2 HEALTH LABELS ... 18
2.3.3 ECONOMIC HEALTH INTERVENTIONS ... 20
2.4 SELF-REGULATION THEORY ... 21
2.4.1 SELF-REGULATION AND HEALTHY SHOPPING DYNAMICS ... 22
2.4.2 SELF-REGULATION AND DRIVERS OF HSD ... 23
2.5 OVERVIEW HYPOTHESES ... 24
3. METHODOLOGY ... 25
3.1 DATA COLLECTION ... 25
3.2 SAMPLE AND CRITERIA SCREENING ... 25
3.3 OPERATIONALIZATION OF VARIABLES ... 26
3.3.1 HEALTH INDICES ... 26
3.3.2 TREND AND VOLATILITY ... 27
3.3.3 DRIVERS OF HSD ... 28
3.4 RESEARCH METHOD ... 29
4. RESULTS ... 32
6 4.1.1 EXPLORATORY ANALYSIS ... 32 4.1.2 MODEL ASSUMPTIONS ... 34 4.1.3 INTERPRETATION ... 35 4.2 MODEL 2 ... 38 4.2.1 EXPLORATORY ANALYSIS ... 38 4.2.2 MODEL ASSUMPTIONS ... 40 4.2.3 INTERPRETATION ... 40 4.3.4 PREDICTIVE VALIDITY ... 43 5. CONCLUSION ... 45 5.1 DISCUSSION ... 46
5.1.1 HEALTHY SHOPPING DECISIONS ... 46
5.1.2 DISCOVERING PATTERNS TO FORECAST DECISIONS ... 47
5.2 LIMITATIONS AND FURTHER RESEARCH... 49
5.3 MANAGERIAL IMPLICATIONS ... 50
5.4 FINAL CONCLUSION ... 51
REFERENCES ... 52
APPENDICES ... 56
APPENDIX 1: DIFFERENCES BETWEEN STORES ... 56
APPENDIX 2: VARIATIE = VOORDEELWEKEN ... 57
APPENDIX 3: MODEL ASSUMPTIONS MODEL 1 ... 57
3.1 MULTICOLLINEARITY ... 57
3.2 NORMALITY ... 57
3.3 HETEROSCEDASTICITY ... 59
APPENDIX 4: MODEL ASSUMPTIONS MODEL 2 ... 59
4.1 MULTICOLLINEARITY ... 59
4.2 NORMALITY ... 59
7
1. INTRODUCTION
Over the past years, several trends in consumer food markets have evolved. In particular, the trend towards ‘healthy choices’ in particular has been quite substantial, as a result of a worldwide problem: obesity. Since 1980, worldwide obesity has doubled, leading to several consequences, such as cardiovascular diseases, diabetes, musculoskeletal disorders and multiple types of cancer (World Health Organization, 2015). Especially in the Netherlands, the rates for obesity are shocking: 40% of the population of 4 years and older is overweight and another 10% of the people is considered to be obese (CBS, 2014). According to Ng et al. (2014), overweight and obesity caused 3 to 4 million deaths, 3-9% of years of life lost, and 3-8% of disability-adjusted life-years in 2010 only. On a large scale, this problem is driven by overconsumption of unhealthy, energy-dense and nutrient-poor foods that have high concentrations of fat, sugar, and salt (Asfaw, 2011). This identifies the need for healthier lifestyles, which can be stimulated by grocery retailers themselves (Payne et al., 2014). In fact, many grocery retailers aim their promotional activities towards stimulating customers to make more conscious, healthy choices. In order to do this more effectively, it would be insightful to know how people shop at a grocery store: how does the healthiness of customers’ shopping behavior play a role and what the sequence is in which they buy products. Unfortunately, there is still little understanding on how this influences the healthiness of the shopping basket.
8
To observe consumer decisions in the most natural way, data from a Dutch grocery retailer will be used. Grocery retailers are one of the most interesting sources for observing consumer behavior for several reasons. First, even though the growth of online stores is a major trend in consumer food markets, still around 50% of all groceries are purchased in brick-and-mortar grocery stores (Glanz et al., 2012). Second, several other trends are ongoing, such as that the recession of the past years has led to a cost-saving-orientation of the shopping public. As a result, shoppers’ stated priorities when choosing their groceries are aimed towards quality, taste or price. Finally, the interest for making healthy choices has grown among customers. However, their actual purchase behaviors do not always seem to actually follow this state of mind.
In this study, a first step will be made to uncover such drivers of (un)healthy shopping behavior based on real purchase data. Plus, a large Dutch grocery retailer, has made data that incorporates the sequence in which customers do their groceries available for this research. Plus is a grocery chain with 255 supermarkets held by 218 entrepreneurs (Plus, 2016a). The fact that all supermarkets are franchisers, makes this retailer a special case compared to other grocery retailers. The entrepreneurs have more freedom when it comes to including for instance local products in the supermarket, while still following the national marketing campaigns. These campaigns incorporate the four brand values that the company stands for: Attention, Quality, Local, and Responsible (Plus, 2015). Recently, Plus received the award for being the best supermarket in promoting Corporate Social Responsibility, making it even more interesting to take Plus as a source to study healthy shopping behavior (GfK, 2016).
1.1 RESEARCH QUESTIONS
This paragraph lists and explains several research questions that guide this study. Even though the importance of finding several drivers of healthy shopping behavior has been mentioned before, the main goal of this research is to uncover what affects customers’ healthy decision-making. The HSD that were mentioned before, could be an example of such drivers. Real-time data should uncover what these dynamics look like in actual data, as up to now the only preliminary results that reflect this are based on experimental data (Van Ittersum and Bijmolt, 2015). When there is an idea of how such dynamics evolve, the effect of HSD of all shopping decisions will be tested on the average health of the basket. Besides, the roles of emotion and self-regulation will be included in this research, as they might explain why certain patterns exist.
9
methods work the best (Giesen et al, 2011; Wansink and Chandon, 2006; Waterlander et al., 2013). In this study, two health interventions that drive healthy shopping behavior will be included: Economic Health Interventions and Health Labels. In collaboration with Plus, two health interventions can be studied. These are not only the specific marketing campaigns of the supermarket that can be called ‘Economic Interventions’ (Variatie = Voordeelweken), but also health labels, which are widely available on products in the Netherlands since the introduction of ‘Het Vinkje’ in 2006 (Het Vinkje, 2016). This label is carried in the food and beverage industry, retail and foodservice and is created to help consumers in making healthier decisions. Beside these two, another driver of shopping behavior is included: ‘General Promotions’. Supermarkets have different price options and many customers do their groceries in a very price conscious manner (Glanz et al., 2012). Therefore, it is expected that such promotions probably also drive the final decision and that price can be chosen over ‘healthiness’.
This all results in the following research question and subquestions:
1. How are healthy shopping decisions influenced during a shopping trip?
1.1 What do HSD look like in real purchases?
1.2 Do HSD affect the average healthiness of decisions throughout the shopping trip?
1.3 Do other drivers, such as health interventions and general promotions affect the average healthiness of decisions throughout the shopping trip?
Addressing these issues will help uncovering what influences healthy choices. However, this still poses a problem, because knowing what drives healthy choices does not necessarily mean that customers’ behaviors are always intentionally influenced. Naturally, we assume that people are rational beings and make conscious choices. That would imply that people’s behavior is then predictable. However, a lot of unplanned buying occurs in the supermarket, which implies that behavior is perhaps not always rational (Gilbride et al., 2015). Therefore, a prediction model will be formed afterwards, to investigate whether people make rational choices and if it is even useful to try and influence people when they are tempted to buy products impulsively. This results in the second research question:
2. Can a pattern be distinguished in the scanner data that can forecast the
healthiness of customers’ purchase decisions?
10
1.2 RELEVANCE
The objective of this study is to discover whether there is such a phenomenon as HSD, how the dynamics of choices during the shopping trip affect the healthiness average level of the basket and if such dynamics can be used to forecast the healthiness of future choices. This will give an indication in how grocery retailers can direct customers towards interchanging unhealthier options for healthier alternatives. Subsequently, the effects of two types of health interventions and of general promotions will be absorbed in the model to investigate whether these have an impact on the HSD and eventually influence the healthiness of the current decision, which gives the possibility to predict the healthiness level of the next purchase.
Up to now, much research has been oriented at discovering what influences the healthiness of product-decisions customers make during the shopping trip. This has shown that for instance the size and shape of the package of a product has an influence (Ordabayeva and Chandon, 2013; Wansink, 1996), as well as the location of healthy and unhealthy food products inside the supermarket (Desai and Ratneshwar, 2003) and prices of these food products (An 2012; Andreyeva, Long and Brownell, 2010) influence the healthiness of single-product purchases. However, there is little to no research performed on subsequent purchases during the shopping trip. So far, only the study of Waterlander et al. (2013) takes the total shopping trip into account. Outcomes of this study are striking: the positive effect of the single product-purchases may be eradicated when taken as part of a larger shopping trip. Additionally, food labels do not have a large effect on food purchases, whereas price discounts do encourage the purchase of healthy products. However, these price cuts do not discourage the purchase of unhealthy foods and lead to larger end-of-trip basket. Therefore, different articles indicate that more research is needed to unravel how pricing strategies can best be designed to result in overall improved food purchases and what role food labels could have to reach this goal. Besides, a pilot study by Van Ittersum and Bijmolt (2015) has shown that there is reason to believe HSD exist.
Taking the previous mentioned studies together, there is enough reason to investigate this matter. A question remains why this has not been investigated in the past. Due to the lack of availability of data of grocery shoppers that take the sequence of shopping into account, specific research to find out how customers make healthy shopping decisions was simply not possible until this point. Recently, supermarkets introduced the options for customers to skip the line and scan their groceries already while doing them using a handscanner. This scanner saves the sequence in which the groceries are into the basket and thus provides the academic world a grand insight in to how customers shop.
11
shopping trips and how active health interventions of the supermarket can stimulate the purchases of healthier food products. With the current consumer trends to eat healthier and the growing problem of obesity, more supermarkets in the Netherlands are developing campaigns that focus on healthier, more responsible purchases (Plus, 2016a). Plus can use the results of this study to stay competitive and continuously keep its customers satisfied.
1.3 OUTLINE
12
2. THEORETICAL FRAMEWORK
This chapter reviews existing literature to propose the underlying hypotheses to answer the research questions. First, the conceptual model is represented in a visual way. Afterwards, numerous paragraphs elaborate on the variables that are pointed out in the models and several hypotheses are presented. Moreover, a number of psychological theories are linked to the relationships in the models to give possible explanations for the relationships in the model. Finally, at the end of the chapter an overview of all hypotheses is provided.
2.1 CONCEPTUAL FRAMEWORK
As shortly explained before, the objective of this study is to uncover the influence of health interventions, promotions and HSD on the average health index of the basket, which is displayed in Fig. 1 by the blue arrows. Subsequently, the aim is to uncover whether such choices are all rational and if a prediction model can be estimated, which is visually displayed with the red variable and red arrows.
13
2.2 HEALTHY SHOPPING DYNAMICS
In the first chapter HSD were already mentioned briefly, but no clear definition of this phenomenon has been stated yet. This paragraph will first discuss different researches that already looked into shopping dynamics in general, followed by a short elaboration on previous work in the field of healthy food purchases, resulting in a definition of the term.
Up to now, the existence of shopping dynamics in general has been proven in different studies, where mainly the contribution of Dhar et al. (2007) is important. They identified that customers go through a process which is called the shopping momentum. This refers to the psychological impulse that is provided when an initial purchase is made and that enhances the purchase of a second, unrelated product. This theory was linked to previous work by Gollwitzer et al. (1990), which explains the occurrence of the shopping momentum as a result of the psychological process caused by the initial purchase. This makes the consumer move from a deliberative to an implemental mind-set, driving subsequent purchases. Dhar et al. (2007) also describe manners in which this shopping momentum can be interrupted.
Beside the shopping momentum, there is also an excessive amount of literature on the phenomenon of impulsive and unplanned purchase behavior. When customers walk through a supermarket, they are confronted with many items that they could potentially purchase, possibly leading to unplanned buying (Gilbride et al., 2015). In this state, two types of dynamics can be distinguished: carryover effects of earlier purchases on subsequent unplanned versus planned purchases, and a change in the probability of making an unplanned versus a planned purchase over the course of the shopping trip (Gilbride et al., 2015). One of the reasons that such impulsive purchases take place is affect, or better said the mood of the consumer at the moment of making the purchase decision (Vohs and Faber, 2007). Several other theories, both social and psychological, could underlie these dynamics (Cannuscio et al., 2014), which will be further discussed in paragraph 2.4.
14
Fig. 2 Healthy shopping Dynamics - results from pilot
study
Source: Van Ittersum and Bijmolt (2015)
Thus, according to this pilot study there is reason to believe that HSD exist. Along this line, the following definition of HSD will be leading throughout this study: ‘HSD are shifts in the healthiness indices of all combined purchase decisions throughout the shopping trip’. The following subparagraphs describe six ways in which HSD can possibly be measured. It needs to be noted that there is a difference between the meaning of the health index that is used in this study and healthiness. An increase in the health index denotes a decrease in the healthiness of the basket/next purchase, since the health index is based on the number of calories. When this number increases, the healthiness thus decreases. All hypotheses will be stated in terms of the health index of the basket or of the next purchase.
2.2.1 HEALTH INDEX OF THE FIRST PURCHASE DECISION
The first experience in a sequence of experiences tends to have a stronger influence on the judgment of individuals than the following experiences, due to primacy effects (Montgomery and Unnava, 2009). An example of such effects is that when people memorize a list of words, they put greater attention on the first words compared to the following ones, resulting in better memorization of them (Greene, 1986). In the topic of healthy choice behavior, this indicates that the healthiness of the first purchase would have a great impact on the following shopping behavior of the customer. This should be taken into account and results in the following hypotheses:
H1A The health index of the first purchase decision is positively related to the average health
index of the basket
H2A The health index of the first purchase decision is positively related to health index of the
next purchase
2.2.2 HEALTH INDEX OF THE PREVIOUS PURCHASE DECISION
15
for recency effects was found in an experiment by Kahneman et al. (1993). They show that people choose to rather feel pain for a longer amount of time, provided that this experience is ended with a pleasant feeling, instead of feeling pain for a shorter amount of time, where this pleasant part is not present. Translating this in the context of healthy shopping behavior, this implies that customers recall the healthiness of their most recent purchase decision more than of the decisions they made before that, making the last purchase decision an interesting factor to take into consideration. Combining this fact with the self-regulation theories on guilt that will be described in paragraph 2.4, it is expected that customers will compensate the relative unhealthiness of their previous purchase with a healthier next choice. This results in the second hypothesis for Model 2:
H2B The health index of the previous purchase decision is negatively related to the health index
of the next purchase
Besides, as compared to primacy effects, recency effects are expected to dominate when affecting the healthiness of the current purchase decision, because recall diminishes when the time since the first decision increases (Greene, 1986). Therefore, the third hypothesis for Model 2 is stated:
H2C The effect of the health index of the previous purchase decision on the health index of the
next purchase decision is larger than the effect of the health index of the first purchase
2.2.3 TREND
A trend of subsequent experiences can either be increasing or decreasing. Consumers usually prefer improvement over a certain amount of time compared to decline, which is called their negative time preference (Loewenstein and Prelec, 1993). In the case of healthy purchase behavior, it can be concluded that customers with an improving trend of healthy choices are more likely to choose a relatively healthy product again than customers with a more negative trend. Even though the pilot study by Van Ittersum and Bijmolt (2015) suggests that the pattern in HSD is not linear, it is still valuable to discover whether healthy shopping behavior improves or declines throughout the shopping trip, which results in the next hypotheses:
H1B An improving trend of healthy choices has a positive influence on the average health index
of the basket
H2D An improving trend of healthy choices has a positive influence on the health index of the
16
2.2.4 PEAKS
During the shopping trip, peak moments in the level of healthiness will occur. Such peaks have an impact on later choice behavior, because the most intensive moments are remembered the best (Montgomery and Unnava, 2009). The same holds for the reversed situation: an extreme ‘low point’ is also remembered more. It does not matter when the healthy/unhealthy peak takes place during the shopping trip. When a very healthy choice is made, this might strengthen the motivation to make healthy decisions again through feelings of pride, or to do the opposite and find the justification to choose unhealthier products, which is called licensing (Khan and Dhar, 2006; Mukhopadhyay and Johar, 2007; Williams and DeSteno, 2008). These concepts are further elaborated upon in paragraph 2.4. Therefore, it is expected that the healthy peak has an influence in both models, but the sign is unknown. For the unhealthy peaks, it is expected that through feeling of guilt customers will tend to make a healthier decision afterwards (Chen and Sengupta, 2014). This concept of guilt will also be discussed later on in paragraph 2.4. This results in the following hypotheses:
H1C Healthy peaks during the shopping trip have an influence on the average health index of
the basket
H1D Unhealthy peaks during the shopping trip have a negative influence on the average health
index of the basket
H2E Healthy peaks during the shopping trip have an influence on the health index of the next
purchase
H2F Unhealthy peaks during the shopping trip have a negative influence on the health index of
the next purchase
2.2.5 VOLATILITY
17
that results in a negative evaluation (Anderson, 2003). Thus, existing literature does not seem to give a clear result of volatility, and applying this to healthy shopping decisions it seems logical that when customers choose many products with very different levels of healthiness, it is harder to predict their next move than for customers who have quite stochastic behavior. Although it is expected that there is some influence, the next hypotheses cannot give a conclusive direction:
H1E The volatility of the health indices of previous purchases influences the average health index
of the basket
H2G The volatility of the health indices of previous purchases influences the health index of the
next purchase
2.2.6 AVERAGE HEALTHINESS OF THE BASKET SO FAR
For Model 2 the dependent variable of Model 1 is included in the model as an additional driver of the next purchase decision. When shopping, customers have orientations that differ from one another. Some of them might go grocery shopping and buy certain items for hedonic reasons, whereas other customers might feel the urge to buy healthy items (Arnold and Reynolds, 2003). The overall healthiness level of the previous purchase decisions that were made could give an indication for people’s intention to buy healthier products. Therefore, when customers already made relatively healthy choices overall before, they must have a higher probability of making a healthier decision again and vice versa. Such a causality closely follows the ideas of self-regulation theory, which describes how people set goals and how they need to control themselves in order to achieve such goals. If customers shop for, on average, healthier groceries over the trip, this may indicate that they will choose healthier products again. More elaboration on this and more accompanying theories will follow in paragraph 2.4. Taking everything in consideration, this results in the following hypothesis:
H2H The average health index of previous purchases has a positive influence on the health index
18
2.3 DRIVERS OF HSD
Now that HSD have clearly been defined and that several indicators have been established, a deeper look is taken into what drives customers’ healthy choice behavior. First, these could be general promotions. Second, this could be health interventions. The importance of health interventions by retailers or suppliers has been investigated in several studies (Giesen et al, 2011; Wansink and Chandon, 2006; Waterlander et al., 2013). They distinguish between different methods in which customers could be guided towards making healthier purchase-decisions. In these articles two main groups of health interventions can be identified: health labels and economic interventions. The first subparagraph describes the effect of general promotions, which are also expected to influence buying behavior. The following two subparagraphs will elaborate on the before mentioned interventions.
2.3.1 GENERAL PROMOTIONS
Every week, supermarkets have different items on promotion. Such promotions have the purpose for customers to make more unplanned purchases, and with success (Inman et al., 1990). In fact, low cognition customers even purchase goods that are on promotion by just the look of the promotion signal, without even checking whether there is a real price discount (Inman et al., 1990). Moreover, research has shown that promotions can accelerate purchases in 2 ways. First, the acceleration of customers’ purchases of the product and second the acceleration of the shopping trip to the store (Kahn and Schmittlein, 1992). Therefore, general promotions in grocery stores do have a large impact. However, it will also result in more unplanned behavior, having a negative impact as customers might lose track of their shopping goal. More theories that can be linked to this are provided in paragraph 2.4. The more products on promotion are added to the basket, the higher the health index of the basket is expected to be, thereby decreasing the healthiness. This results in the following two hypotheses:
H1F General promotions have a positive influence on the average health index of the basket H2I General promotions have a positive influence on the health index of the next purchase 2.3.2 HEALTH LABELS
19
consumers compared to people with a normal weight. Their results show that low-fat labels lead to overconsumption of snack foods by all consumers, but that these effects are stronger for people who already are overweight. Besides, the presence of salient serving-size information such as ‘Contains two Servings’ reduces overeating for people with a normal weight, but has no impact on overweight consumers. Therefore, this research thus indicates that food labels do have an impact on consumers. However, the people that are overweight are not paying enough attention to such health labels. To stimulate this to a greater extent, manufacturers and retailers could consider making labels more explicit by altering the packages or promoting these characteristics more heavily. However, other work by Waterlander et al. (2013) on health labels and pricing strategies to influence healthy shopping behavior gives different results. The outcomes of this study show that price effects overshadow the effects of food labels. These food labels per se do not have any significant effect on the purchase of healthy foods.
In this study, the effectiveness of ‘Het Vinkje’, which was shortly mentioned before, will be investigated. There are two types of Vinkjes as shown in Fig. 3: one with a green and another with a blue circle (Het Vinkje, 2016).
Fig. 3 Het Vinkje
The green logo indicates that the food product belongs to the healthier products of the food pyramid and contains important nutrients that you need on a daily basis. The blue logo indicates that the product does not belong to the food pyramid and that you should not eat this too often, but that it is a better choice within the product category. The effectiveness of the Vinkje is shown in an internal research by Plus (2016b), which indicates that 83% of the customers are aware of the Vinkjes and 65% experiences it as a positive addition. The case is, however, that only 19% of the customers actively pays attention to the Vinkjes while doing groceries.
Recently, the largest Dutch customers association ‘De Consumentenbond’ started a campaign against this health label. According to them, Dutch consumers are not well aware of what the two labels mean and according to them it does not result in healthier choices (Consumentenbond, 2016). This is all based on a qualitative research among 1057 panel members. An interesting fact about these researches about the Vinkjes is that they are based on questionnaires that were filled out by a panel. There is, however, no known research that investigates the Vinkjes in a quantitative way. Therefore, this study might add different insights to the effects of this health label.
20
This reflects the effect of the health label itself, not of the healthiness of the products that carry it. It does not necessarily have to concern a very healthy product, but as the blue labels indicated it can also be placed on a product that is a better choice within a relatively unhealthy product category. This results in the following hypotheses:
H1G Health labels have an influence on the average health index of the basket H2J Health labels have an influence on the health index of the next purchase 2.3.3 ECONOMIC HEALTH INTERVENTIONS
Mainly due to economic shocks such as a recession, falling income or dramatic increases of food prices, purchase behavior can be influenced (Andreyeva et al., 2010). Times like these create pressure to purchase food that is lowest in cost, making processed, unhealthier foods more attractive. In theory, there would be two ways to deal with situations like these and stimulate healthier purchase-behavior: either lowering prices of healthy food products (i.e. a subsidy), or raising prices of relatively unhealthy products (i.e. a fat tax). Different studies already indicated that mainly the first intervention, a subsidy on healthier products, could be a successful way to stimulate healthy shopping. According to An (2012), subsidizing healthier foods tends to be effective in modifying dietary behavior. The only constraint to this finding is that long-term effectiveness and impact on the overall diet intake are unknown.
Waterlander et al. (2012) studied the effects of price subsidies and taxes on respectively healthy and unhealthy foods throughout the entire shopping trip. They found that price increases on unhealthy food products up to 25% of the original price do not result in differences in healthy food purchases. This indicates that the tendency to purchase healthier food products will only increase when a substantial tax on unhealthy food is introduced. Besides, their results showed that price discounts on healthy foods have two effects. On the one hand they encourage customers to purchase healthy products. On the other hand it makes customers increase the energy of the total shopping basket, resulting in an equally (un)healthy shopping basket. This indicates that the complete purchase process of customers is more dynamic and is not only explained by prices.
There thus seems to be a positive influence of economic interventions when single-product purchases are made, but no change on the healthiness of the complete basket. Therefore, it is expected that there is some influence of economic health interventions in both models, but no clear cut direction of that relationship. This results in the following hypotheses:
21
2.4 SELF-REGULATION THEORY
This paragraph discusses different psychological theories that interfere with how customers make decisions in a grocery store. This study focuses on the decisions customers make during a shopping trip and there are many possible influences that can distract them. The paragraph continues by describing several underlying mechanisms that possibly explain why customers make certain choices, in combination with the variables that were described previously in the conceptual model.
When making the decision to buy a healthy product, a certain level of self-regulation is required. Baumeister and Heatherton (1996) describe three ingredients of this self-regulation. First,
standards are important, which are ideals, goals or other conceptions of possible states. These
standards are essential, as either a dilemma of conflicting standards or even a lack of having any obstructs effective self-regulation. Second, monitoring entails the current state of being that is compared to the standard and loops of feedback of one’s actions which are necessary to guide an individual to their goals. When people cease to follow their actions, they tend to lose control. The third and last phase is called operate, which follows the second phase closely. If it turns out that the current state is not compatible with the standards, a certain process is set in motion to change this. The first two ingredients have been researched widely, but it is quite unsure how these processes in the last phase actually function as they seem to be much more complex.
22
2.4.1 SELF-REGULATION AND HEALTHY SHOPPING DYNAMICS
Some of the theories described in the previous paragraph can be closely linked to the decision process that is researched in this study: is an individual going to choose something they might like better in terms of taste, but is unhealthy, or can they regulate their actions and choose for healthier products? It goes without saying that customers need to have the internal goal to make such decisions. When customers simply do not care about eating healthier, this theory cannot be applied.
Since the publication of the article of Baumeister and Heatherton (1996), many researchers followed this up with studies on self-regulation theory and depletion of the self-regulatory resources. In more recent work, an opposite phenomenon is found with regard to the self-regulation resource. When two consecutive self-regulatory situations require similar control processes, the self-regulation resource does not get exhausted, but in fact enhances (DeWitte et al., 2009).
Beside exercising this control, customers also feel more subjective emotions during the shopping trip. Different concepts such as licensing, pride and guilt can be linked to the process that is captured within HSD. These three concepts all have their influence in different ways, but affect the choice for healthy products in a positive way. First, licensing occurs when “a prior, virtuous intent boosts people’s self-concepts, thus reducing negative self-attributions associated with the purchase of relative luxuries” (Khan and Dhar, 2006, p. 256). In the current research context, this means that when customers have to motivation to make a healthy choice, this boosts their self-concepts and results in a them feeling that it is justified to make a second, unhealthier choice. Second, pride can play a part within the shopping trip. When a consumer made the choice to purchase a healthy product instead of a relatively unhealthy one, he or she resisted and facilitated self-regulation, which will give a sense of pride (Mukhopadhyay and Johar, 2007; Williams and DeSteno, 2008). Third, guilt plays its part too when customers are unable to resist the temptation of buying a relatively unhealthy product, making them feel more motivated to continue their shopping trip with the purchase of a healthier product (Chen and Sengupta, 2014).
In addition to these three concepts that boost healthy shopping behavior, there is one final concept that needs to be discussed briefly. It is possible that customers continue to buy unhealthy foods after their first failure, which is called What-The-Hell (Cochran and Tesser, 1996). The name itself already reflects that this behavior cannot be explained by any of the previous theories and is thus merely observed. All of these concepts will most likely be identified within the data that is made available by Plus.
23
determined by elements of the current trip and previous shopping trips. Their findings indicate that the probability for unplanned behavior increases as the shopping trip continues.
Since even research does not seem to agree on how self-regulatory processes develop during a series of choices, it is hard to predict HSD. In the next subparagraph, a closer look will be taken into the interconnection of self-regulation theory and drivers of HSD.
2.4.2 SELF-REGULATION AND DRIVERS OF HSD
It is also a possibility to connect self-regulation theory to health interventions, as these interventions aim to influence customers’ choice behavior and general promotions. Not a lot of research within self-regulation is performed with regard to promotions within the store. At the start of their shopping trip, most customers have some idea of what they want to buy, but shopping goals might be a bit fuzzy (Lee and Ariely, 2006). As the trip proceeds, these goals become clearer. However, promotions influence customers spending more when their goals are less concrete compared to customers with less fuzzy goals (Lee and Ariely, 2006). In general, promotions already seem to influence behavior more than health labels (Waterlander, 2013). Thus the influence of promotions is expected to be larger than the influence of health labels in influencing healthy shopping behavior, resulting in the last hypotheses:
H3A The effect of (healthy) promotions is larger than the effect of health labels in influencing
healthy shopping decisions in Model 1
H3B The effect of (healthy) promotions is larger than the effect of health labels in influencing
healthy shopping decisions in Model 2
24
2.5 OVERVIEW HYPOTHESES
Table 1 Overview of all hypotheses
Hypotheses +/-
H1A The health index of the first purchase decision is positively related to the average health index of
the basket
+
H1B An improving trend of healthy choices has a positive influence on the average health index of the
basket
+
H1C Healthy peaks during the shopping trip have an influence on the average health index of the basket +/-
H1D Unhealthy peaks during the shopping trip have a negative influence on the average health index of
the basket
-
H1E The volatility of the health indices of previous purchases influences the average health index of the
basket
+/-
H1F General promotions have a positive influence on the average health index of the basket +
H1G Health labels have an influence on the average health index of the basket +/-
H1H Economic health interventions have an influence on the average health index of the basket +/-
H2A The health index of the first purchase decision is positively related to health index of the next
purchase
+
H2B The health index of the previous purchase decision is negatively related to the health index of the
next purchase
-
H2C The effect of the health index of the previous purchase decision on the health index of the next
purchase decision is larger than the effect of the health index of the first purchase
H2D An improving trend of healthy choices has a positive influence on the health index of the next
purchase
+
H2E Healthy peaks during the shopping trip have an influence on the health index of the next purchase +/-
H2F Unhealthy peaks during the shopping trip have a negative influence on the health index of the next
purchase
-
H2G The volatility of the health indices of previous purchases influences the health index of the next
purchase
+/-
H2H The average health index of previous purchases has a positive influence on the health index of the
next purchase
+
H2I General promotions have a positive influence on the health index of the next purchase +
H2J Health labels have an influence on the health index of the next purchase +/-
H2K Economic health interventions have an influence on the health index of the next purchase +/-
H3A The effect of (healthy) promotions is larger than the effect of health labels in influencing healthy
shopping decisions in Model 1
H3B The effect of (healthy) promotions is larger than the effect of health labels in influencing healthy
25
3. METHODOLOGY
3.1 DATA COLLECTION
The data used in this study is made available by Plus, one of the largest Dutch grocery retailers. Since the introduction of their new store concept a few years ago, Plus provided the availability of self-scanning devices for customers in many grocery stores. What makes this data unique from regular scanner data is that it saves the sequence in which customers scanned the items as they walked through the grocery store. In this manner, the way customers grocery shop and the sequence in which they make decisions can be traced precisely.
The data was collected in January and February of 2016 and contained one pre-promotion week, one promotion week and one post-promotion week for three different Plus stores. These three different stores are similar in size, but differ in the type of neighbourhood where they are located. To test whether there are any differences between supermarkets, a one-way ANOVA and three regressions were performed (Appendix 1). These tests indicate that there are no major differences between the three stores. Therefore, the customers that form three groups divided over the different stores are combined in one data file to look at the total customer database available for this research. The three weeks for which there is data available (i.e. the pre-promotion, promotion and post-promotion week) were used to test the previously stated hypotheses.
In addition to this scanner data, the internal data of Plus on the nutritional information of their products was also used to create insights about the healthiness phenomena. The nutritional information was, however, far from complete and thus calorie information for all missing products had to be found on product catalogues from other retailers, blogs and articles available online. When this file was complete, it was merged with the scanner data file. Afterwards, the calories were recalculated and transformed into indices. Paragraph 3.3 provides more detail on the operationalization of the healthiness of the products.
3.2 SAMPLE AND CRITERIA SCREENING
26
1. HSD are expected to be found with different choices. When a customer chose the same product a number of times in a row, this was observed as one choice they made. The aggregation of all baskets resulted in smaller baskets for almost all customers;
2. These smaller baskets were then categorized in remaining sizes. All baskets that contained less than 10 different products were deleted from the dataset. This way the dataset is based on customers with shopping baskets that were formed through a substantial amount of choices.
As a result, the remaining baskets all contained at least 10 products without any duplicates. Subsequently, due to issues with the missing nutritional information, it was necessary to take subsamples from every store for every week. Samples of 300 customers were drawn, which resulted in a final sample of 2.700 customers.
3.3 OPERATIONALIZATION OF VARIABLES
In this sub-paragraph, the operationalization of several of variables in the model will be described.
3.3.1 HEALTH INDICES
To measure the healthiness of purchase decisions, the nutritional information of the products that customers bought is used. For customers, this information is available at the back of each product in the supermarket, containing information about for instance sugar, calories, salt and carbs. The importance of such information becomes clear in the research by Burton et al. (2006), where customers filled out a survey and participated in an experiment that aimed to uncover how well customers are aware of the amounts of fat and calories. Their results specify that a shocking amount of customers is not aware of the high amount of calories and fat in the food they consume. This illustrates the added value of nutritional information, as it can have a positive impact on public health and should therefore not be overlooked. Other researches also uncover the important added value that calorie information has on customers’ awareness of the (un)healthiness of the food they consume as well (Giesen et al., 2011). In this study, the number of calories that a product contains were used as an indicator of the healthiness of that product. The amount of calories per 100 grams that is always provided on the package of the product was used. Not only because this gives an equal idea of the relative (un)healthiness of products with in their category, but also because this is the information that customers have available when shopping at the grocery store.
27
category, an index <1 implies that the product is relatively healthy, and an index >1 implies that it is relatively unhealthy. Besides, the average healthiness of each product category is calculated by taking the average amount of calories of each product category, and creating health indices for every product category. Then the health index of each product is multiplied by the health index of the product category it belongs to. This operationalizes what healthy choices are and how they evolve over a shopping trip. To illustrate this a bit more clearly, the process of creating the health indices is described with the following formulas in two steps:
1. Average number of calories of the product category =Total number of calories within the product category
Total number of products in the category
2. Health Index of the product = Number of calories of Product J
Average number of calories of the product category (1)
The second formula shows the final health index that was used for each product. This health index adjusts the healthiness of the product for the healthiness of the product category it belongs to.
Moreover, the calculated health indices under (2) were used as a basis to operationalize the variables for the primacy and recency effects and the healthy and unhealthy peaks:
- Primacy effect = First adjusted Health Index (2) for each customer i - Recency effect = Previous adjusted Health Index (2) for each customer i
- Healthy peak = Lowest/minimum adjusted Health Index (2) for each customer i - Unhealthy peak = Highest/maximum adjusted Health Index (2) for each customer i
Finally, the dependent variable ‘average health index’ indicates whether the customer chose relatively more healthier or unhealthier products during the trip. This variable is calculated by
𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝐻𝑒𝑎𝑙𝑡ℎ 𝐼𝑛𝑑𝑒𝑥𝑖=
𝑇𝑜𝑡𝑎𝑙 𝑠𝑢𝑚 𝑜𝑓 ℎ𝑒𝑎𝑙𝑡ℎ 𝑖𝑛𝑑𝑖𝑐𝑒𝑠 𝑜𝑓 𝑎𝑙𝑙 𝑝𝑟𝑜𝑑𝑢𝑐𝑡 𝑐ℎ𝑜𝑖𝑐𝑒𝑠 𝑏𝑦 𝑐𝑢𝑠𝑡𝑜𝑚𝑒𝑟 𝑖 𝑇𝑜𝑡𝑎𝑙 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑑 𝑏𝑦 𝑐𝑢𝑠𝑡𝑜𝑚𝑒𝑟 𝑖
3.3.2 TREND AND VOLATILITY
For every customer in the dataset, a trend line is calculated which describes the linear trend for the health indices (HI) for all t moments during the shopping trip:
HI = ß0 + ß1 t + Ɛt (1)
28
as well and their method for calculating this will be closely followed in this study. In order to find out how volatile the health indices are from choice to choice, the autocorrelation of the error terms of Eq. (1) can be used. Autocorrelation can arise in two ways: positive or negative. Positive autocorrelation refers to the occurrence of residuals in t that have the same sign as the residual in t-1. Negative autocorrelation shows a pattern of positive and negative values for the residuals compared to the trend line interchanging (Leeflang et al., 2015). In the case of health indices, it is expected that healthy and unhealthy choices will interchange a lot, resulting in a curved patterns, which was shown before in Fig. 2. Therefore, negative autocorrelation would reflect this the best way. As a measure to evaluate this autocorrelation, the Durbin-Watson statistic is used. This statistic ranges from 0 to 4, where a value close to 0 indicates positive autocorrelation, close to 2 indicates non-autocorrelation and close to 4 indicates negative autocorrelation (Shehu et al., 2016). Therefore, a higher value of the Durbin-Watson statistic indicates higher volatility. When following the method performed in the study by Shehu et al. (2016), it becomes clear that in this dataset the average Durbin-Watson statistic is 1,6812, which indicates that most of the patterns show slight positive autocorrelation. This implies that most cases do not show strong variability.
3.3.3 DRIVERS OF HSD
29
3.4 RESEARCH METHOD
Since the data used for this study is of a numerical form, the research methods used are quantitative. As stated before, there are two objectives to this research. The first objective is testing how healthy shopping decisions are influenced during the shopping trip. This is Model 1, for which the drivers were stated in the first conceptual model. The equation below describes the regression model for the first model. It needs to be mentioned that the Durbin-Watson statistic will only be included in the regression when the average value indicates that there is in fact autocorrelation. This is tested in the next chapter. Model 1 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝐻𝐼𝑖𝑗 = 𝛽0+ 𝛽1 𝐹𝑖𝑟𝑠𝑡𝑖𝑗 + 𝛽2 𝐻𝑒𝑎𝑙𝑡ℎ𝑃𝑒𝑎𝑘𝑖+ 𝛽3 𝑈𝑛ℎ𝑒𝑎𝑙𝑡ℎ𝑃𝑒𝑎𝑘𝑖+ 𝛽4 𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖 + 𝛽5 𝑇𝑟𝑒𝑛𝑑𝑖+ 𝛽6 𝐺𝑒𝑛𝑒𝑟𝑎𝑙 𝑃𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛𝑖+ 𝛽7 𝐻𝑒𝑎𝑙𝑡ℎ 𝐿𝑎𝑏𝑒𝑙𝑠𝑖+ 𝛽8 𝐸𝑐𝑜𝑛𝑜𝑚𝑖𝑐 𝐻𝑒𝑎𝑙𝑡ℎ 𝐼𝑛𝑡𝑒𝑟𝑣𝑒𝑛𝑡𝑖𝑜𝑛 + 𝜀𝑖𝑗 Where
𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝐻𝐼𝑖𝑗 = The average health index of product j for customer i of all
products in the final basket, which is calculated by 𝐻𝐼1+ 𝐻𝐼2+ 𝐻𝐼3+⋯+ 𝐻𝐼𝑗−1
𝑗−1 for each customer
𝐹𝑖𝑟𝑠𝑡𝑖𝑗 = The health index of the first product j chosen by customer i
𝐻𝑒𝑎𝑙𝑡ℎ𝑃𝑒𝑎𝑘𝑖 = The healthy peak in one of the health index of the products
in the basket of customer i
𝑈𝑛ℎ𝑒𝑎𝑙𝑡ℎ𝑃𝑒𝑎𝑘𝑖 = The unhealthy peak in one of the health index of the
products in the basket of customer i
𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖 = The variability of the health indices in all the choices made
by customer i
𝑇𝑟𝑒𝑛𝑑𝑖 = The slope of trend line of all health indices of the products
bought by customer i
𝐺𝑒𝑛𝑒𝑟𝑎𝑙 𝑃𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛𝑖 = The amount of products on promotion bought by customer i
𝐻𝑒𝑎𝑙𝑡ℎ 𝐿𝑎𝑏𝑒𝑙𝑠𝑖 = The amount of products containing a health label bought
by customer i
𝐸𝑐𝑜𝑛𝑜𝑚𝑖𝑐 𝐻𝑒𝑎𝑙𝑡ℎ 𝐼𝑛𝑡𝑒𝑟𝑣𝑒𝑛𝑡𝑖𝑜𝑛𝑖 = Dummy (0/1) that indicates whether the products were
30
Model 2 will be presented after an explanation of the operationalization of the second dependent variable: health index of the next purchase. The objective of Model 2 is to find out whether a pattern can be distinguished in the healthy shopping decisions customers make while shopping for groceries. One product category will be taken into account to test whether it is possible to build a prediction model that forecasts how relatively healthy the next purchase decision in this category will be. In theory, this could be attempted for all product categories in the supermarket. However, to try whether it works, it will first only be built for one category. All categories have different healthiness scores relative to the other categories, and some of them have larger differences between the health scores of products within the category than others. In this study, the product category ‘dairy’ is chosen to test the prediction model, for the following reasons:
1. When looking at the store plan of the average Plus supermarket, it becomes evident that the dairy section is situated in the second part of the store. This means that customers already made a number of choices before getting to this section, which provides the basis for building a prediction model;
2. Dairy is usually perceived as quite a healthy product, as it contains calcium which is good for the body and can prevent for instance osteoporosis at a later age (Voedingscentrum, 2016). The data also shows that compared to other product categories, dairy is a relatively very healthy category, with a health index far below one (0,3933). As this study investigates healthy shopping behavior, it makes sense to choose a category that is healthy and to see whether the healthy choices can be predicted;
3. Even though that on average the dairy section contains relatively healthy products, within the category itself the decisions can vary from very healthy (Lowest Health Index = 0,130) to very unhealthy (Highest Health Index = 5,547). Therefore, even while the decision to buy a dairy product per se is healthy, within the section the choice can still be possibly unhealthy.
31
products. These results were compared to the actual results and a naïve model and several tests were performed to show the predictive validity of the model.
Model 2
As the conceptual model in paragraph 2.1 described already, one variable will be added to capture the recency effect: ‘Previous’. Another variable that is added to the model is the ‘average HI of the basket so far’. Model 1 investigated the drivers of this variable. This results in the following final outline for Model 2:
𝐻𝐼𝑖𝐽 = 𝛽0+ 𝛽1 𝐹𝑖𝑟𝑠𝑡𝑖𝑗+ 𝛽2 𝑃𝑟𝑒𝑣𝑖𝑜𝑢𝑠𝑖𝑗+ 𝛽3 𝑃𝑜𝑠𝑃𝑒𝑎𝑘𝑖+ 𝛽4 𝑁𝑒𝑔𝑃𝑒𝑎𝑘𝑖+ 𝛽5 𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖+
𝛽6 𝑇𝑟𝑒𝑛𝑑 + 𝛽7 𝐺𝑒𝑛𝑒𝑟𝑎𝑙 𝑃𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛𝑖+ 𝛽8 𝐻𝑒𝑎𝑙𝑡ℎ 𝐿𝑎𝑏𝑒𝑙𝑠𝑖+
𝛽9 𝐸𝑐𝑜𝑛𝑜𝑚𝑖𝑐 𝐻𝑒𝑎𝑙𝑡ℎ 𝐼𝑛𝑡𝑒𝑟𝑣𝑒𝑛𝑡𝑖𝑜𝑛 + 𝛽10 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝐻𝑒𝑎𝑙𝑡ℎ𝑖𝑛𝑒𝑠𝑠 𝐵𝑎𝑠𝑘𝑒𝑡𝑖+ 𝜀𝑖𝑗
𝐻𝐼𝑖𝐽 = Health index of the next product purchased J
𝐹𝑖𝑟𝑠𝑡𝑖𝑗 = The health index of the first product j chosen by customer i
𝑃𝑟𝑒𝑣𝑖𝑜𝑢𝑠𝑖𝑗 = The health index of the previous product j chosen by
customer i
𝐻𝑒𝑎𝑙𝑡ℎ𝑃𝑒𝑎𝑘𝑖 = The healthy peak in one of the health index of the products
in the basket of customer i
𝑈𝑛ℎ𝑒𝑎𝑙𝑡ℎ𝑃𝑒𝑎𝑘𝑖 = The unhealthy peak in one of the health index of the
products in the basket of customer i
𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖 = The variability of the health indices in all the choices made
by customer i
𝑇𝑟𝑒𝑛𝑑𝑖 = The slope of trend line of all health indices of the products
bought by customer i
𝐺𝑒𝑛𝑒𝑟𝑎𝑙 𝑃𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛𝑖 = The amount of products on promotion bought by customer i
𝐻𝑒𝑎𝑙𝑡ℎ 𝐿𝑎𝑏𝑒𝑙𝑠𝑖 = The amount of products containing a health label bought
by customer i
𝐸𝑐𝑜𝑛𝑜𝑚𝑖𝑐 𝐻𝑒𝑎𝑙𝑡ℎ 𝐼𝑛𝑡𝑒𝑟𝑣𝑒𝑛𝑡𝑖𝑜𝑛𝑖 = Dummy (0/1) that indicates whether the products were
bought during a VVW or not by customer i
𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝐻𝑒𝑎𝑙𝑡ℎ𝑖𝑛𝑒𝑠𝑠 𝐵𝑎𝑠𝑘𝑒𝑡𝑖 = The average healthiness of the basket up to now compiled
32
4. RESULTS
4.1 MODEL 1
4.1.1 EXPLORATORY ANALYSIS
In order to obtain some preliminary insights about the data, a correlation matrix was created for all the independent and dependent variables of Model 1. As Pearson’s correlation matrix in Table 2 shows, there are multiple significant correlations between the variables. A possible explanation for these correlations is that the first 6 variables in the list (Health Index, Volatility, HealthPeak, UnhealthPeak, First and Trend) are all based on the health index. This seems to be quite a plausible explanation, as the general promotions and health labels do not seem to correlate with the health index. Whether these correlations result in problems such as multicollinearity is tested in paragraph 4.2.
Table 2 Pearson’s Correlation matrix Model 1
Average Health Index
First HealthPeak UnhealthPeak Volatility Trend Promotions Health Labels Economic Health Intervention Average Health Index 1 ,176** ,197** ,409** -,063** ,032 ,072** ,126** ,046* First ,176** 1 ,033 ,088** -,095** -,083** -,020 -,002 -,008 HealthPeak ,197** ,033 1 -,088** ,135** ,071** ,004 -,026 ,004 UnhealthPeak ,409** ,088** -,088** 1 -,044* ,015 -,056** ,072** ,046* Volatility -,063** -,095** ,135** -,044* 1 -,017 -,040* -,016 -,025 Trend ,032 -,083** ,071** ,015 -,017 1 ,033 ,023 ,019 Promotions ,072** -,020 ,004 -,056** -,040* ,033 1 ,015 ,127** Health Labels ,126** -,002 -,026 ,072** -,016 ,023 ,015 1 -,032 Economic Health Intervention ,046* -,008 ,004 ,046* -,025 ,019 ,127** -,032 1
* indicates that the correlation is significant at the 5% level (2-tailed) ** indicates that the correlation is significant at the1% level (2-tailed)
33
matter in which sequence the categories are visited. The average amount of products purchased during all shopping trips is 18 items. Therefore, only these and an additional number of items in the sequence are displayed in Fig. 4.
Fig. 4 Healthy Shopping Dynamics
Note: <1 = relatively healthy, 1= relatively neutral and >1 = relatively unhealthy.
The trend line resembles the ‘rollercoaster’ pattern that was mentioned before. It shows that customers usually start their shopping trip with a relatively unhealthy trend. Afterwards, from item 5 until item 17, customers seem to pick products that are relatively healthy (<1) for a very long time. Then, it seems that the pattern changes and trend of the shopping behavior becomes quite unhealthy. Given that this graph was made with the shopping behavior of approximately 2.700 customers, it can be said that this pattern is quite robust. The fluctuations in the figure prove that HSD can be distinguished in real purchases. If there was a flat line around 1, this would indicate that customers all shop very differently and that the average sequence of all customers would add up the same for each point in the trip. The fact that there are a lot of fluctuations throughout the shopping trip, indicates that HSD do evolve.
34
Table 3 Independent Samples T-test for economic health intervention
Economic Health Interventions
Economic Health Intervention Mean St. Deviation Average HI 0 ,9957 ,17311
1 1,0123 ,18492
Independent Samples T-Test
Levene’s Test for Equality of Variances
F Significance T-value DF Significance
Equal variances not
assumed (α < 0,10) ,001 ,076* -2,244 1694,77 0,025
As Table 3 shows, the difference between the shopping weeks is significantly different. What is striking, is that the average health index of the baskets is lower in weeks where there is no economic health intervention, and thus the baskets are healthier in those weeks. This is not the expected effect, considering that the promotional weeks are focusing on getting customers to choose healthier products. Whether this effect will also come through in the regression analysis, is identified in paragraph 4.1.3.
Before finding out whether any of the drivers are statistically significant and interpreting the beta’s of the variables of the regression model, the model assumptions need to be checked. In the case that there are violations of these assumptions, remedies might change the betas and their corresponding levels of significance (Leeflang et al., 2015).
4.1.2 MODEL ASSUMPTIONS
35
plot of the unstandardized residuals of that model is visualized, which indicates that these residuals indeed seem to be normally distributed (Appendix 3.2).
Third, to test for autocorrelation the Durbin-Watson statistic was calculated once again. This result (DW = 1,991) indicates that there is no reason to assume that autocorrelation plays a part, as a value around 2 indicates that there is no autocorrelation (Leeflang et al., 2015).
Finally, heteroscedasticity was tested . In order to do so, the differences between the healthy promotion weeks (VVW) and regular weeks was taken into account. Levene’s test for equality of variances was performed, where the null hypothesis states that there is equality of variances and thus homoscedasticity. In case this hypothesis is rejected, there is a problem with the residuals. Since this is not the case, this problem does not exist in this model (Appendix 3.3).
Since none of the model assumptions were violated in a sense that the regression output changed, the results can be interpreted without re-estimating the model.
4.1.3 INTERPRETATION
In this paragraph the regression model results (Table 4) can be interpreted.
Table 4 Results Regression Model 1
Model Statistics
Model F-value 533,038 R2 0,613
Model Significance 0,001** Adjusted R2 0,612
Regression Output
Beta Std. Error T-value Significance
Constant ,654 ,010 63,705 ,001** First HI ,043 ,004 9,602 ,001** Healthy Peak ,305 ,011 27,012 ,001** Unhealthy Peak ,075 ,001 57,379 ,001** Volatility ,018 ,004 4,403 ,001** Trend ,360 ,061 5,897 ,001** Promotions ,017 ,021 ,837 ,403 Health Labels ,021 ,022 ,967 ,334
Economic Health Intervention ,001 ,005 ,322 ,747
** indicates significance at the 1% level (2-tailed)
