CONSTRAINING THE OBESITY EPIDEMIC BY UNDERSTANDING THE ROLE OF HEALTHY SHOPPING DYNAMICS

(1)

CONSTRAINING THE OBESITY

EPIDEMIC BY UNDERSTANDING THE

ROLE OF HEALTHY SHOPPING

DYNAMICS

by Sanne Schreurs

(2)

2

Master Thesis

CONSTRAINING THE OBESITY EPIDEMIC BY

UNDERSTANDING THE ROLE OF HEALTHY

SHOPPING DYNAMICS

Sanne Schreurs University of Groningen Faculty of Economics and Business

MSc. Marketing Management and Marketing Intelligence 15-01-2017 Het Gemaal 13 8604 DN Sneek +31 (0) 681956907 sanneschreurs@outlook.com S2810662

(3)

3

Management summary

Over the past decades, only a few health topics have caused as much concern and debate as obesity. Despite the efforts to prevent and manage obesity, researchers expect the obesity rates to increase by 130% over the next two decades. The obesity epidemic is largely driven by the overconsumption of unhealthy foods. Given that over half of all food purchases are occurring at supermarkets, supermarkets can play an important role in stimulating healthy purchase behavior.

Current research has focused on the healthiness of single-product purchases. However, the results of a pilot study indicate that there are certain patterns in the way consumers make (un)healthy choices while grocery shopping. These patterns are called: healthy shopping dynamics and are expected to be driven by an interdependence between the healthiness of sequential purchases, meaning that when customers make an initial healthy decisions they license themselves to make an unhealthy choice next. Hence, current research is insufficient and even counterproductive in curbing the obesity epidemic. The purpose of this research is therefore to understand the role of healthy shopping dynamics during larger shopping trips. In order to investigate this issue, we had the unique opportunity to collaborate with a large Dutch grocery retailer: Plus. Retailer Plus shared basket level data captured by hand-held scanners from customers. The final dataset used contained 5400 shopping trips of customers covering week 1/6 of 2016 in three different Plus stores. Two models were

estimated in this paper. The first model was a random effects panel regression to identify the drivers that are of influence on the healthiness of the next product purchase. The second model was a linear regression to understand the drivers of the healthiness of the end of trip basket.

(4)

4

With respect to the drivers of healthy shopping dynamics the study suggest that the salience of blue health labels does not have an impact on the healthiness of the next product purchase or the healthiness of the end of trip basket. The salience of green health labels does have an impact, however contrary to expectations this effect is negative.

Lastly, some control variables (i.e. week, time of the day) were included in this study to test their potential effects. Interesting is that the results show some preliminary evidence of a 'good intentions' effect in week 1 of 2016, where customers have strong health intentions to start the year healthy. Moreover, the results indicate some preliminary evidence of a positive effect of the health intervention held by retailer Plus (variation = benefits weeks). Hence, it seems that customers are showing relatively more healthy purchase behavior in week 2 as supposed to other weeks. Lastly, the results show that customers make relatively more healthy choices in the morning (09:00-12:00) as supposed to later times of the day (12:01-21:00).

(5)

5

Preface

My passion for marketing started three years ago during my two internships for my bachelor commercial economics. I worked for Valess (vegetarian products) and Campina, both brands of the company FrieslandCampina. During both internships I was a junior brand manager and had the opportunity to learn about all the facets of retail marketing from inspiring persons. After finishing my bachelor, I realized that there was so much more to learn about Marketing. Moreover, my supervisors advised me to start a master in Marketing since I had the ambition to work in the FMCG industry. Therefore, I started my pre-master Marketing in September 2014. After a year of hard work, I was ready to start the Master Marketing in September 2015. Soon I realized that I really liked working with big data, something that I would never have expected from myself. Therefore, I decided to study two profiles: Marketing Management and Marketing Intelligence. I really enjoyed studying both tracks and learned a lot. Moreover, I developed a passion for data analytics which I never have expected in the first place. Therefore, I am very grateful that I got the opportunity to work with big data from Plus for this master thesis, since it suits both of my passions: FMCG and data analytics. So I would really like to thank my first supervisor dr. ir. K. van Ittersum and second supervisor prof. dr. T.H.A. Bijmolt for the opportunity to work on this interesting project. Moreover, I would like to thank them for their thorough supervision and the valuable discussions. Last but not least, I would like to thank Plus for sharing the data and Astrid Westerveld MSc, for her support during my research.

Sanne Schreurs

(6)

1 Introduction

Over the past decades, only a few health topics have caused as much concern and debate as obesity. Globally, the number of overweight and obese individuals has increased from 921 million in 1980 to 2.1 billion in 2013 (Ng et al. 2013). Being obese is a health problem since it is an important risk factor for many chronic diseases, certain types of cancer and untimely death (Hruby et al. 2016, Tomer 2011). Overweight and obesity were globally estimated to cause 3.4 million deaths, result in a loss of 3.9% years of life and lead to 3.8% of disability adjusted life years in 2010. Besides the dramatic health risks, obesity also has an huge impact on health care costs. Research suggests that almost 10% of all medical spending is caused by obesity, leading to costs of $147 billion per year (Finkelstein et al. 2012, Finkelstein et al. 2009). Despite the efforts to prevent and manage obesity, researchers expect the rates to even increase by 130% over the next two decades (Fiechtner et al. 2016, Finkelstein et al. 2012, Hruby et al. 2016, Ng et al. 2014).

Shephard (2006) argues that weight gain occurs by repeatedly eating and drinking more calories than needed to generate energy for daily activities. Hence, the obesity epidemic is largely driven by the overconsumption of unhealthy, energy-dense and nutrient-poor foods which are high in sugar, salt and (saturated) fat (Asfaw, 2011). Given that over half of all food purchases are occurring in supermarkets, supermarkets are facing public pressure to address the issue of obesity (Glanz et al. 2012, Regmi and Gehlhar 2005). In an attempt to approach the problem supermarkets established nutritional information on their products (Fischer et al. 2011, Sutherland et al. 2010), included healthier product offerings, tried to educate shoppers (Glanz et al. 2012) and reduced the prices of healthy food (Cohen et al. 2015, Seymour et al. 2004). Despite these efforts, there is still little evidence of improvement. Some of these attempts even backfired. Whereas, reducing the prices of healthy foods leads to an improved healthiness of single product purchases, the total healthiness of the end of trip basket actually decreases (Waterlander et al. 2012, Waterlander et al. 2013a).

(9)

9

2003), packaging (Ordabayeva and Chandon 2013, Wansink 1996), and health labels (Wansink and Chandon 2006a) on the healthiness of single-product purchases, these effects may be eliminated during larger shopping trips (Waterlander et al. 2012, Waterlander et al. 2013a). Hence, the preliminary results of a pilot study by Van Ittersum and Bijmolt (2015) suggest that certain patterns exist in the way consumers make (un)healthy choices while grocery shopping. These patterns are called: healthy shopping dynamics and represent the patterns in the healthiness of sequential purchase decisions. These patterns are expected to be driven by an interdependence between the healthiness of sequential purchases, meaning that when customers make an initial healthy decision they can license themselves to make an unhealthy choice next (Wilcox et al. 2009). Moreover, this interdependence is expected to be driven by the different shopping emotions that consumers experience while grocery shopping (Chen and Sengupta 2014, Cochran and Tesser 1996, Mukhopadhyay and Johar 2007, Wilcox et al. 2009).

Obviously, these healthy shopping dynamics may be one of the keys in constraining the obesity epidemic. However, at the moment only preliminary evidence of healthy shopping dynamics exists (Van Ittersum and Bijmolt 2015). Therefore, this paper will investigate these healthy shopping dynamics and attempts to understand how the healthiness of the end of trip

basket comes about. More specifically, two models will be developed in this paper. The first

model will focus on identifying the drivers that are of influence on the healthiness of the next product purchase. The second model will focus on how the healthiness of the end of trip basket comes about. In order to investigate these issues, we had an unique opportunity to collaborate with a large Dutch grocery retailer: Plus. Retailer Plus is interested to participate in this research since it is positioned as a social responsible grocery retailer and therefore has a focus on local, sustainable and fair-trade food (Plus 2016). Moreover, Plus is actively stimulating consumers to make healthier choices by executing different health interventions. Plus shared basket-level data captured by hand-held scanners from customers. Therefore, the exact sequence in which customers purchased their groceries in their basket is known. This data is expanded with ingredient data in order to determine the healthiness of each purchase. The final dataset used, consists in total of 5400 shopping trips of customers, covering week 1/6 of 2016 in 3 different stores.

This paper extends current research in multiple ways. First, this is the first study that

(10)

10

data including the sequence of purchases. Therefore, it is possible to investigate the presence and influence of healthy shopping dynamics in real-life, thereby observing consumers decisions in the most standard way. Second, this study recognizes that grocery shopping is a dynamic process in which shoppers have to make multiple product decisions which are a sequence of interdependent choices. Therefore, this research goes beyond current insufficient research which only focuses on the healthiness of single-product purchases. Third, this study will generate insights on how the healthiness of the end of trip basket comes about. Next to academic relevance, this research is also managerial relevant since it provides insights in how grocery retailers can direct customers towards healthier alternatives.

Research questions:

1. What are the drivers of the healthiness of the next product purchase? 2. What are the drivers of the healthiness of the end of trip basket?

(11)

11

2 Theoretical framework

This chapter starts with presenting the conceptual model. The subsections will

review existing literature and present the underlying hypothesis regarding the variables in the conceptual model. This will lead to an overall understanding of the phenomenon. The chapter will end with an overview of all hypothesis.

2.1 Conceptual model

Please refer to Figure 1 for the conceptual model.

2.2 Healthy shopping dynamics

(12)

12

pain of paying. Pain of paying is defined as "the notion that a consumer who pays for a

product or service experiences emotions associated with the act of paying" (Zellermayer 1996, p.6). Hence, they show that consumers experience an immediate pain of paying when making a purchase, which has an influence on latter purchase decisions. Building on this literature Sheehan and Van Ittersum (2012) demonstrate and explain why shoppers' relative spending decreases early in the shopping trip and increases again during the second half of the shopping trip. They demonstrate a concave pattern in shoppers' relative spending throughout a shopping trip.

Based on the research above it can be concluded that the grocery shopping process is more than a sequence of independent purchase decisions. Hence, the grocery shopping process is a dynamic process which evolves in response to earlier purchase decisions.

Building on these shopping dynamics literature, Van Ittersum and Bijmolt (2015) proposed to investigate whether such a pattern is also present with regards to (un)healthy purchase

behavior during major shopping trips. In order to investigate whether these so called patterns:

healthy-shopping dynamics are present during major shopping trips, they conducted a pilot

study. Healthy shopping dynamics are defined as: "systematic and nonlinear patterns in the healthiness of consecutive purchases that are driven by an interdependence between the healthiness of these purchases" (Van Ittersum 2016, p.2). The results of this pilot study reveal some preliminary evidence for the presence of healthy shopping dynamics. More specifically, the findings suggest that the relative healthiness evolves non-linearly throughout a series of purchase decisions. However, since this pilot study is the only research investigating the presence of healthy shopping dynamics, the evidence is scarce. The following subparagraphs will describe the psychological theories related to healthy shopping dynamics.

2.3 Psychological theories relating to healthy shopping dynamics

As argued, the results of a pilot study by Van Ittersum and Bijmolt (2015) reveal some preliminary evidence of healthy shopping dynamics. More specifically, the study shows that the relative healthiness evolves non-linearly throughout a series of purchase decisions. The relative healthiness of a product is in this study calculated by dividing the amount of calories a product contains (calories per 100 gram) by the average amount of calories of the

(13)

13

help to better understand the idea and reasoning behind the existence of healthy shopping dynamics. The next subsections will discuss these different theories in order to better understand healthy shopping dynamics.

2.3.1 Attitudes, goals, intentions and behavior

Consumers' purchase decisions are influenced by a variety of factors. One of these factors is consumer's attitudes. Attitudes can be defined as: "the psychological tendency that is

expressed by evaluating a particular entity with some degree of favor or disfavor" (Eagly and Chaiken 1993, p.1). Research suggests that consumers' attitudes towards products are major determinants of buying behavior (Eagly and Chaiken 1993). According to research by Eagly and Chaiken (2007) attitudes are formed based on three evaluative responses: cognitive, affective and behavioral evaluative responses. First, when people are evaluating an attitude object, they hold certain beliefs. These beliefs are called the cognitive evaluative responses which range from very positive to very negative. Whereas some people belief that fast food is tasty, other people think about the amount of calories and consequences for their health. Second, when evaluating an attitude object people experience certain feelings, moods and emotions which can be labeled as affective evaluating responses. Hence, some people will experience feelings of joy when eating a piece of pie, but others will feel guilt because of the bad consequences for their health. Lastly, behavioral evaluative responses are about the intention to act or actions that people perform with respect to the attitude object.

Research suggests that attitudes are not directly influencing purchase behavior, but that this effect is mediated by intention (Ajzen 1991, Fishbein and Ajzen 1977, Sheppard et al. 1988, Webb and Sheeran 2007a). Hence, the theory of reasoned action is the most influential theory in this area (Fishbein and Ajzen 1977). This theory suggests that the intention to perform specific behavior (i.e. eating healthy) is determined by an individual's attitudes towards the specific behavior and subjective norms. According to this theory an Figure 2: healthy shopping dynamics - results from a

(14)

14

individual is more likely to perform a specific behavior when the individual holds positive attitudes towards the behavior and when the behavior is expected by people important to the person (Fishbein and Ajzen 1977). However, Sheppard et al. (1998) argue that this theory is only focused on behavior (i.e. eating healthy) instead of goals that result from that behavior (i.e. losing weight). Goals can be defined as cognitive representations of desired end-states (Fennis and Stroebe 2015). According to Cochran and Tesser (1996) a goal is a source of motivation to perform certain behavior. Therefore, this theory was extended to the theory of planned behavior by adding a third component: perceived behavioral control. Perceived behavioral control can be defined as the extent to which performing a specific behavior is under an individual's control (Ajzen 1991). According to this theory people take their own resources and capabilities into account when forming intentions to perform certain behavior.

When relating these theories to healthy shopping behavior it can be concluded that people's attitudes (formed based on three evaluative responses) partly determine the intention to buy healthy products. Hence, when individuals hold positive attitudes towards healthy food and perceive that buying this healthy food is under their control, they will form intentions to buy healthy products when grocery shopping. Forming these implementation intentions will lead to remembering the action intention when the situation arises and therefore increase the likelihood that individuals act in line with these intentions (Webb and Sheeran 2007a). 2.3.2 Self-control theory

Attitudes thus partly determine consumers' intention to buy healthy products. However, acting accordingly to these intentions requires exerting self-control.

The ultimate goal of many people is to increase their personal well-being (Ryan and Deci 2000). Since health is seen as an important determinant of well-being, people are nowadays more sensitive to issues concerning their health (Moorman and Matulich 1993). Hence, people feature a basic health motivation which activates them to pursue health goals (Moorman and Matulich 1993). In order to reach those health goals people need to exert self-control, which can be defined as conscious self-regulation (Muraven and Baumeister 2000). The purpose of exerting self-control is to maximize the long-term interest of the individual (Kanfer and Karoly 1972, Mischel 1996). For example rejecting a piece of pie because one's on a diet requires self-control. According to Baumeister & Heatherton (1996) there are three phases to self-control. First, clear standards (e.g. ideals, goals) are needed, such as the

(15)

15

state of the self to the standards. When people quit monitoring their actions and states, they lose control. The third phase is named operate, when the current state is evaluated negatively to the standards set, a process will be embedded to change the current stage.

Based on the literature review above, the question arises how it can be that obesity and overweight rates continue to rise when people are able to exert self-control? There are multiple answers to this question:

First, although there are many different forms of self control they all draw on a common resource: self control strength. This self control strength is a limited resource and therefore readily depleted (Muraven et al. 1998). When exerting self control, it consumes control strength, which diminishing the amount of strength available for following self-control attempts (Muraven and Baumeister 2000). Thus, after initial self-self-control, consumers are likely to fail at subsequent self-control attempts (Vohs et al. 2005).

Second, people have to find a balance between maximizing pleasure and minimizing harm when it comes to their health (Rabia et al. 2006). When people are confronted with a temptation, a motivational conflict may arise which can be defined as a conflict between a wish for a desired object and people's other goals (e.g. to lose weight) (Giner-Sorolla 2001). Since self control is a limited resource, the likelihood of giving in to the temptation increases after each initial self-control attempt (Muraven et al. 1998, Muraven and Baumeister 2000, Vohs et al. 2005).

2.3.3 Shopping emotions

(16)

16

Consumers that just made a healthy product choice followed by one or more healthy

purchases (exerting self-control) may experience feelings of pride (Mukhopadhyay and Johar 2007). However, following an initial healthy product choice consumers could also license themselves to make an unhealthy choice next. Hence, when individuals have acted in line with a long-term goal they sometimes license themselves to pander to temptations (Wilcox et al. 2009). Thus, if people focus on their progress towards a goal, they sometimes allow themselves to temporarily disengage from that goal in order to pursue indulgent options (Fishbach and Dhar 2005).

Shoppers make a(n)...

Following a.... ...Healthy choice ...Unhealthy choice

....Healthy choice Pride License to indulge

....Unhealthy choice Guilt What the hell (WTH)

Table 1 : shopping emotions

In addition there are two shopping emotions at work after an initial unhealthy product choice. Consumers that made an unhealthy product choice followed by one or more unhealthy

purchases may experience feelings of guilt (Chen and Sengupta 2014). However, buying one or multiple unhealthy products after an initial unhealthy product could also lead to the:

what-the-hell effect. This effect is mostly seen when individuals eat more calories on a day than

they are allowed to. Since these individuals perceive that the day is lost, they see no more incentive to constrain calorie intake and instead overindulge (Cochran and Tesser 1996).

According to research pride and guilt are self-conscious emotions, meaning that these emotions are found in social relationships and arise from concerns about others' evaluation of the self. Individuals deeply care about maintaining a favorable view of the self (Cialdini and Goldstein 2004). Whereas a negative evaluation of the self or the behavior of the self is fundamental to guilt, a positive evaluation of the self leads to pride (Eisenberg 2000, Haidt 2003, Tangney 1993). Negative self-conscious emotions will promote moral behavior (Eisenberg 2000, Haidt 2003). However, this is not always the case since repeatedly buying unhealthy products could also lead to the what-the-hell effect instead of moral behavior.

(17)

17

eat/don't eat conflict. Third, individuals think irrational about their diet. Hence, individuals perceive the caloric limit in daily terms so that if the daily limit is exceeded there is no necessity for further restraint for the subgoal (i.e. a day).

Overall it can be concluded that consumers experience different emotions while grocery shopping. These emotions are expected to influence the relative healthiness throughout a series of purchase decisions.

2.4 Capturing Healthy shopping dynamics

The previous paragraphs explained the healthy shopping dynamics and the psychological theories behind this phenomena. This paragraphs will explain how the healthy shopping dynamics are going to be measured.

2.4.1 Relative healthiness

According to Shephard (2006) weight gain occurs by repeatedly eating and drinking more calories than needed to generate energy for daily activities. Hence, consumers considerably underestimate the amount of calories, especially for high-calorie products (Burton et al. 2006). It is estimated that men on average need 2500 calories a day and woman need 2000 calories a day. However, when people want to lose weight, they should consume on average 500 to 700 fewer calories a day (Voedingscentrum 2016). Research suggests that consumers obtain around two-third of their calories from food prepared at home (Kozup et al. 2003). Therefore, reducing the amount of unhealthy purchases at the point of purchase offers the potential of reducing overconsumption of unhealthy foods and their associated calories.

In this research the relative healthiness of a single product purchase will be based on the amount of calories per 100 gram. More specifically, the 'relative healthiness' for a product can be calculated by subtracting the amount of calories a product contains (per 100 gram) from the average amount of calories in the corresponding product category (baseline). Therefore, a product is relatively healthy when it contains less calories than the baseline. Chapter 3 will provide a more in-depth understanding of the relative healthiness of a single product purchase.

2.4.2 Purchase decisions

Now a days, consumers have to make a lot of decisions. More importantly, these decisions have to be made after consumers already made other choices or judgments. Research suggests that these prior decisions and choices have an influence on latter decisions and choices (Dhar et al. 2007, Khan and Dhar 2006, Waterlander et al. 2012, Waterlander et al. 2013a).

(18)

18

decisions. Since people are striving for personal well-being, they feature a basic health motivation to stay healthy when grocery shopping (Moorman and Matulich 1993). Hence, people develop positive attitudes towards a healthy lifestyle and form implementation intentions to buy healthy products when grocery shopping (Webb and Sheeran 2007a). In order to act in line with these intentions and reach those health goals people need to exert self-control (Muraven and Baumeister 2000). However, self-self-control strength is a limited resource (Muraven et al. 1998). Thus, after exerting self-control, consumers are likely to fail at

subsequent self-control attempts (Vohs et al. 2005). When relating this to healthy choice behavior, it can be argued that an individual has to make multiple decisions while grocery shopping, resulting in exerting self-control multiple times. Therefore, the number of

purchases an individual has made can be used as a proxy for self-control. Since self-control strength is limited, it is expected that an individual is more likely to fail in exerting self-control after multiple decisions. Hence, multiple purchases could therefore result in unhealthy purchases. This results in the following hypotheses:

H1a: The number of purchases (self-control) is negatively related to the healthiness of the next

product purchase.

H1b: The number of purchases (self-control) is negatively related to the healthiness of the end

of trip basket.

2.4.3 Cumulative sum of the healthiness and unhealthiness since an inflection point Self-control is not the only factor expected to influence the healthiness of the next product purchase and end of trip basket. Besides exerting self-control consumers are expected to be influenced by different shopping emotions.

Research on choices among individual outcomes at different points in time suggests that decision makers prefer to have a good outcome sooner rather than later, which is called

positive time preferences (Benzion et al. 1989, Chapman and Elstein 1995, Thaler and Shefrin

(19)

19

results in a series of losses compared to the reference point (Loewenstein and Sicherman 1991, Loewenstein and Prelec 1993). Overall, decision makers prefer an improvement over a certain amount of time, e.g. series of gains (Chapman 2000). When relating this to healthy choice behavior it can be argued that consumers with an improving sequence of healthy purchases are more likely to choose a relative healthy product again. However, a pilot study by Van Ittersum and Bijmolt (2015) suggests a systematic and non-linear pattern in the healthiness of consecutive purchases. Therefore, research is contradicting.

This discrepancy between patterns may be explained by the shopping emotions consumers experience during grocery shopping. It is expected that there are four different shopping emotions at work, two that a consumer could experience after initial healthy purchases and two after initial unhealthy purchases (Chen and Sengupta 2014, Cochran and Tesser 1996, Mukhopadhyay and Johar 2007, Wilcox et al. 2009). Consumers that made one or more healthy purchases after an initial healthy purchase could experience feelings of pride and therefore keep on choosing relatively healthy products next (Mukhopadhyay and Johar 2007). However, when consumers keep on making healthy purchases, the cumulative sum of the healthiness in their shopping trip keeps on increasing. Therefore, it could also be that there is a certain tipping point where consumers will use their healthy progress, to license

themselves to indulge and choose a relatively unhealthy product next (Wilcox et al. 2009). When consumers choose to indulge at a certain point in time, the relative healthiness will change from positive to negative. This change from relatively positive to relatively negative is called a negative inflection point (Van Ittersum 2016). Although it is argued that consumers could experience pride after an initial healthy choice, it is expected that multiple healthy choices will result in an unhealthy next product purchase. Hence, when consumers are making consecutive relatively healthy choices, the cumulative sum of the healthiness

increases for that consumer, leading to an increased chance on a negative inflection point (license to indulge) and therefore a relative unhealthy next purchase. This results in the following hypothesis:

H2a: the cumulative sum of the healthiness since a positive inflection point has a negative

effect on the healthiness of the next purchase.

(20)

20

However, when consumers keep on making unhealthy purchases, the cumulative sum of the unhealthiness increases during their shopping trip. Therefore, it could also be that consumers experience feelings of guilt at a certain point in time (Chen and Sengupta 2014). These

feelings of guilt are expected to lead to a positive inflection point and therefore a healthy next product purchase (Chen and Sengupta 2014). Thus, we speak of a positive inflection point when the relative healthiness of consecutive product purchases is changing from relatively unhealthy to relatively healthy (Van Ittersum 2016). Although it is argued that consumers could experience the what-the-hell effect after initial unhealthy choices, it is expected that multiple unhealthy choices will result in a healthy next product purchase. Hence, when consumers are making consecutive relatively unhealthy choices, the cumulative sum of the unhealthiness increases for that consumer, leading to an increased chance on a positive inflection point (feelings of guilt) and therefore a relative healthy next product choice. This results in the following hypothesis:

H2b: the cumulative sum of the unhealthiness since a negative inflection point has a positive

effect on the healthiness of the next purchase.

As argued consumers are likely to fail at subsequent control attempts after exerting self-control (Vohs et al. 2005). Therefore, it can be argued that the effect of the cumulative sum of the healthiness on the healthiness of the next product purchase is even more negative at the end of the shopping trip. Therefore,

H3a: the number of purchases (self-control) negatively moderates the effect of the cumulative

sum of the healthiness since a positive inflection point on the healthiness of the next purchase.

Moreover, it can also be argued that the effect of the cumulative sum of the unhealthiness on the healthiness of the next product purchase is less positive at the end of the shopping trip. Therefore,

H3b: the number of purchases (self-control) negatively moderates the effect of the cumulative

sum of the unhealthiness since a negative inflection point on the healthiness of the next purchase.

2.5 Other drivers of healthy shopping dynamics

2.5.1 Health labels

(21)

21

Ng et al. 2014, Tomer 2011). Therefore, agencies try to come up with interventions in order to tackle obesity (McCormick and Stone 2007). Whereas, supermarkets made multiple efforts to reduce obesity (Cohen et al. 2015, Fischer et al. 2011, Glanz et al. 2012, Seymour et al. 2004, Sutherland et al. 2010), governments have made very slow progress in the implementation of intervention strategies (Rashad and Grossman 2004).

However, a well-known attempt to educate shoppers about healthy choices are the health labels. The purpose of these health labels is to inform consumers if the product is a healthy choice or not. In the Netherlands there are two types of health labels, please refer to Figure 3. The green health label is present on products that belong to basic nutrition like: bread, vegetables and dairy and indicates a 'healthy choice' within these categories. The blue health label is present on products that are consumed as 'extra' like snacks, sauces and candy and indicates the better choice within these categories (Consumentenbond 2016b).

In March 2016 'de Consumentenbond', a large Consumers Association in the

Netherlands, started a campaign called: 'weg met het vinkje' (remove the health labels). The Consumers Association started this campaign because they believe that the health labels are misleading (Consumentenbond 2016c). There are multiple reasons for this believe. First, the health labels are not mandatory, therefore not all manufacturers are participating in putting a health label on their products. Manufactures for example argue that they don't want a national logo on their international package. Moreover, manufacturers that do want a health label on their products, need to pay a small fee in order to realize this. However, not all manufactures are prepared to pay this fee. This means that it could be that some healthy products in a category do not contain health labels because the manufacturer does not want to participate, which is misleading for consumers (Consumentenbond 2016a). Second, research by the Consumers Association suggests that only 10% of the consumers can correctly explain the difference between the health labels (van der Meulen 2016). Hence, most consumers don't understand the correct meaning of the health labels and therefore misinterpret the meaning of the health labels (Consumentenbond 2016c). In August 2016 it was therefore decided that the blue health label should be removed from the products and that the green health labels should be in line with 'de schijf van vijf' (which contains 5 categories that consists of only healthy products).

(22)

22

Academic research is inconclusive about the effectiveness of health labels. Several

researchers found that attempts to inform shoppers about the healthiness of a certain products e.g. traffic light nutrition labeling and health labels have no direct influence on healthy purchase behavior (Sacks et al. 2011, Waterlander et al. 2013b). However, Wansing and Chandon (2006) provided three studies and found that health labels do have an impact on consumer's consumption behavior. The findings of the first two studies suggests that the presence of low-fat nutrition labels actually increases food-intake. Hence, labeling products as low fat leads to: (1) an increase in food intake by up to 50% and (2) a decrease in

consumption guilt. Interesting is that the findings differ between normal-weight people and overweight people. Whereas for normal-weight people low-fat labels increases consumption of foods that are believed to be relatively healthy, low-fat labels increase the consumption of all foods for people that are already overweighed. Lastly, the third study indicates that salient information about serving sizes (e.g. this plate contains two servings) only reduces overeating for normal weight consumers and not for overweight consumers. Therefore, it could be that people with overweight do notice the health labels but do not necessarily buy these relatively healthier products. Taking the results of these three studies together, it could be argued that health labels have a salient effect (Wansink and Chandon 2006b). Hence, salient stimuli draw attention because the stimulus is noticeably different from the environment (Fennis and Stroebe 2015). When relating this to healthy choice behavior it could be argued that when a consumer is making a purchase in a category where a lot of products contain a health label, this will create salience for healthy shopping behavior and therefore leads to a healthy product purchase. This leads to the following hypotheses:

H4a: The salience of blue health labels in the product category where a consumers is about to

make a purchase has a positive influence on the healthiness of the next product purchase.

H4b: The salience of green health labels in the product category where a consumers is about

to make a purchase has a positive influence on the healthiness of the next product purchase.

(23)

23

H5a: the number of purchases (self-control) positively moderates the effect of the salience of

blue health labels in the product category where a consumers is about to make a purchase on the healthiness of the next product.

H5b: the number of purchases (self-control) positively moderates the effect of the salience of

green health labels in the product category where a consumers is about to make a purchase on the healthiness of the next product.

Lastly, it could be argued that when a consumer has bought products in product categories which contain a lot of health labels, this also has lead to salience for healthy shopping behavior resulting in a more healthy end of trip basket. Therefore,

H6a: The salience of blue health labels during the shopping trip has a positive influence on the

healthiness of the end of trip basket.

H6b: The salience of green health labels during the shopping trip has a positive influence on

the healthiness of the end of trip basket.

2.5.2 Non-food products

Now a day's, supermarkets are not only offering food products but also provide consumers with a wide assortment of non-food products. Whereas this offers convenience for consumers, it offers sales opportunities for supermarkets. The fact that supermarkets are now also offering non food products means that consumers are buying non food products after initial food products and vice versa. As argued, consumers are comparing each point in a sequence to the previous period which than serves as a reference point for decision making (Loewenstein and Sicherman 1991, Loewenstein and Prelec 1993). Therefore, it can be argued that when a customer bought a non food product this becomes the reference point instead of the healthy or unhealthy purchases he/she made. This could in turn lead to a kind of 'reset' so the consumer pays more attention to the healthiness of the next purchase. Therefore,

H7a: if the previous product bought was a non-food product, this has a positive effect on the

healthiness of the next purchase.

Moreover, it could be argued that consumers who bought multiple non food products in a grocery shopping trip, experienced more 'reset' moments which eventually will lead to more healthy purchase behavior. Therefore,

H7b: the higher the percentage of non-food products in a basket, the more healthy the end of

(24)

24

2.6 Overview of hypothesis

Please refer to Table 2 for an overview of the hypothesis.

Hypothesis +/-

H1a The number of purchases (self-control) is negatively related to the healthiness of the next product purchase.

-

H1b The number of purchases (self-control) is negatively related to the healthiness of the end of trip basket.

-

H2a The cumulative sum of the healthiness since a positive inflection point has a negative effect on the healthiness of the next purchase.

-

H2b The cumulative sum of the unhealthiness since a negative inflection point has a positive effect on the healthiness of the next purchase.

+

H3a The number of purchases (self-control) negatively moderates the effect of the cumulative sum of the healthiness since a positive inflection point on the healthiness of the next purchase.

+

H3b The number of purchases (self-control) negatively moderates the effect of the cumulative sum of the unhealthiness since a negative inflection point on the healthiness of the next purchase.

+

H4a The salience of blue health labels in the product category where a consumers is about to make a purchase has a positive influence on the healthiness of the next product purchase.

+

H4b The salience of green health labels in the product category where a consumers is about to make a purchase has a positive influence on the healthiness of the next product purchase.

+

H5a The number of purchases (self-control) positively moderates the effect of the salience of blue health labels in the product category where a consumers is about to make a purchase on the healthiness of the next product.

-

H5b The number of purchases (self-control) positively moderates the effect of the salience of green health labels in the product category where a consumers is about to make a purchase on the healthiness of the next product.

-

H6a The salience of blue health labels during a shopping trip has a positive influence on the healthiness of the end of trip basket.

+

H6b The salience of green health labels during a shopping trip has a positive influence on the healthiness of the end of trip basket.

+

H7a If the previous product bought was a non-food product, this has a positive effect on the healthiness of the next purchase.

+

H7b The higher the percentage of non-food products in a basket, the more healthy the end of trip basket.

(25)

25

3 Methodology

3.1 Data collection and description

The data resides from retailer Plus, one of the largest grocery retailers in the Netherlands. Plus shared basket-level data captured by hand-held scanners from customers. Therefore, the exact sequence in which customers purchased their groceries in their basket is known. The data contains observations for three different stores during the first six weeks of 2016. Moreover, Plus also shared nutritional information on a product level (amount of calories per 100 gram). This information was used in order to generate insights regarding the healthiness of a single product purchases. Unfortunately, this data was far from complete. Therefore, the amount of calories for the products where the information was missing, had to be found via product catalogues from other retailers and blogs. After completing this calorie file, it was merged to the shopper scanner data. This resulted in a dataset that contains both the exact sequence in which customers purchased their groceries and the corresponding amount of calories per product. In this way, the variables regarding the healthiness of a single product purchase could be calculated.

Two other researchers already worked with a portion of the data namely: week 1, week 3 and week 6. Since it is interesting to observe what happens over time, it was decided to expand this portion of data by adding week 2,4 and 5. Therefore, the final dataset contains week 1 until week 6 of three different Plus stores.

There are some interesting events in the data. First, week 1 is included where customers are expected to have strong health intentions, since December (the month full of Holidays e.g. Christmas and New Years Eve) is officially over. This is called 'the good intentions effect'. Figure 4 graphically displays some evidence about this assumption and shows the amount of times people searched on the word 'calories' in Google over time. The highest peak is in the first week of January, meaning that most people searched on the word 'calories' that week. After January the amount of search terms slowly levels off.

(26)

26

Second, since Plus is positioned as a social responsible grocery retailer, Plus actively

stimulates consumers to make healthier choices by executing health interventions (Plus 2016). The data contains such a health intervention which took place in week 2 and 3 of 2016, named 'variatie = voordeel weken' (variation = benefit weeks). During this health intervention

consumers could choose a variety of healthy foods for a discounted price (i.e. choose five different vegetables for €5,- or buy 3 different pieces of fruit and only pay for 2). Therefore, it is able to investigate the influence of such a real-life health intervention on customer's

(un)healthy purchase behavior. The data contains one pre-promotion week (week 1), two promotion weeks (week 2, week 3) and three post-promotion weeks (week 4, 5 and 6). Lastly, Plus shared information about which products contained a green or blue health label. Therefore, it was possible to investigate whether this type of health intervention has an effect on the (un)healthy choice behavior of customers.

3.2 Sample

The shopper scanner data file initially contained 22.773 shopping trips in three different stores for six weeks. These shopping trips were very diverse with respect to the amount of products bought. Whereas some baskets contained only a few products, other baskets contained more than hundreds of products. However, as argued prior decisions and choices have an influence on latter decisions and choices (Dhar et al. 2007, Khan and Dhar 2006, Waterlander et al. 2012, Waterlander et al. 2013a). Therefore, healthy shopping dynamics are expected to arise when consumers have to make multiple different decisions when grocery shopping. In order to allow healthy shopping dynamics to occur the following baskets were included in the sample:

1. Baskets involving unique choices. As argued healthy shopping dynamics are expected to arise with different product choices. Buying multiple items of the same product is therefore seen as one product choice. In order to control for this, a column 'quantity' was added to the data which indicated how often the exact same product was bought (based on barcode and product description). After adding this column, the data was aggregated (removing all lines with the exact same product purchases per consumer). This aggregation resulted in smaller baskets sizes for almost all consumers.

2. Baskets involving 10 or more purchases. As argued healthy shopping dynamics are expected to arise during larger shopping trips that include multiple different product

(27)

27

was added to the data which contained the amount of unique product purchases a customer made. Please note that non food product choices were also seen as a choice.

The criteria above resulted in a shopper data file containing baskets with 10 or more unique product purchases. As argued the nutritional data shared by Plus was incomplete. Therefore, the two other researchers decided to take subsamples from the overall dataset. They used a sample of 300 unique customers per week, per store, which resulted in a total database of 2700 customers. In order to be consistent, this research expanded this database by adding subsamples of 300 unique customers per week per store for week 2,4 and 5. This resulted in a rich database containing 5400 shopping trips of customers, covering week 1 until week 6 of 2016 in 3 different stores.

3.3 Variables

The following subparagraphs will give a detailed description of the different variables and how they are operationalized.

3.3.1 Healthiness of a single product purchase and the end of trip basket It can be argued that at the most basic level, overweight and obesity are driven by the

overconsumption of calories (Shelley 2012, Shephard 2006, Swinburn et al. 2011). Research suggests that consumers obtain around two-third of their calories from food prepared at home (Kozup et al. 2003). Therefore, reducing the amount of unhealthy purchases at the point of purchase, offers the potential of reducing overconsumption of unhealthy foods and their associated calories. However, research suggests that consumers considerably underestimate the amount of calories, especially for high-calorie products (Burton et al. 2006). Therefore, the nutrition labels on packed foods in supermarkets were dramatically changed by the 1990 Nutrition Labeling and Education Act (NLEA). This law required a new label format, namely the Nutritional Facts panel, for displaying nutritional information more prominently

(Balasubramanian and Cole 2002). Since, 13 December 2016, it is compulsory for all food products to contain nutritional information per 100 gram (i.e. calories, amount of fat, sugar, salt) (Ministerie van economische zaken 2016). It is expected that this law will reduce the trend of obesity, since this nutritional information provides information that can assists consumers in making more healthy food choices (Fischer et al. 2011, Sutherland et al. 2010).

Based on the information above it was decided to determine the healthiness of a single-product purchase based on the number of calories. In order to be able to compare the

(28)

28

the number of calories per 100 gram. More specifically, the healthiness of a single product purchase is calculated by: subtracting the number of calories (per 100 gram) of product purchase k of individual i

(x

ik

)

from the average number of calories (per 100 gram) in its

relevant product category j

(

j

)

In order to operationalize the average number of calories in a

product category, the average number of calories of all unique choice options within a product category will be used as a baseline.

In order to illustrate this process more clearly, please refer to the following formula's: 1. Baseline product category j:

_j=

2. Healthiness of a single product purchase:

(

_j

-x

_ik

)

Where:

i= individual i

j= product category j k= product purchase k

The first formula is used to calculate the baseline (average amount of calories) of a specific product category j. The data contains a total of 39 different product categories and each product category has its own baseline. The second formula is used to calculate the healthiness of a single product purchase, by distracting the amount of calories the product contains from the baseline of the product category where the product is located. Therefore, a positive number indicates a relatively healthy choice and a negative number indicates a relatively unhealthy choice.

The dependent variable of model 1 is the healthiness of a single product purchase, calculated based on the method described above. The dependent variable of model 2 is the healthiness of the end of trip basket which is calculated by adding up the healthiness of all products bought by a customer during a shopping trip.

3.3.2 Cumulative sum (un)healthiness since an inflection point

(29)

29

customer experienced a positive inflection point, the lagged cumulative sum of the healthiness started to add up (while the cumulative sum of the unhealthiness stayed 0) until a customer experienced a negative inflection point. After experiencing a negative inflection point, the lagged cumulative sum of the unhealthiness started to add up (while the cumulative sum of the healthiness stayed 0) until a customer experienced a positive inflection point again. 3.3.3 Number of purchases (self-control)

The self-control variable for model 1 is operationalized as the total amount of purchases (including non food purchases) a customer did at time t. Please note that this variable starts counting at 0. Therefore, the number of purchases =0 has a certain meaning, making mean centering unnecessary.

The self-control variable for model 2 is operationalized as the total amount of purchases a customer did during a shopping trip. Please note that the amount of non food purchases was excluded from this variable, since model 2 contains the percentage of non food purchases as a separate variable.

3.3.4 Health labels

In order to determine the effectiveness of blue and green health labels on the healthiness of the next product purchase (model 1) and on the healthiness of the end of trip basket (model 2) the percentage blue and green health labels per category were calculated. This was done by dividing the amount of products with a blue or green health label in a product category by the total amount of products in that product category (calculated for blue and green health labels separately). Hence, this percentage indicates the salience of the health labels in a specific product category. For model 1, the percentage green and blue health labels for the category in which the customer is about to make a purchase is used. For model 2, the weighted average of green and blue health labels was calculated for each customer, which therefore indicates to what extend a customer was exposed to blue and green health labels during a shopping trip. 3.3.5 Non Food

(30)

30

3.3.6 Overview of the variables for model 1 and model 2

Please refer to Table 3 for an overview of the main variables in model 1 and to Table 4 for an overview of the main variables in model 2.

Model 1 Variable name Variable explanation Scale

Dependent variable

Healthiness A variable indicating the healthiness of a single-product purchase at time t.

Ratio

(continuous)

Panel ID variable

UniqueID An unique ID for each shopping trip. N.A.

Time variable Purchase number A variable indicating the corresponding

purchase number within a shopping trip. Starting at 1 until a customer ended the shopping trip. N.A. Independent variables Cumulative sum of the healthiness since IP (Lagged, t-1)

The lagged cumulative sum of the healthiness since a positive inflection point until a negative inflection points takes place. Ratio (continuous) Cumulative sum of the unhealthiness since IP (Lagged, t-1)

The lagged cumulative sum of the unhealthiness since a negative inflection point until a positive inflection point takes place.

Ratio

(continuous)

Nthpurchase (self-control)

The total amount of purchases at time t, including the non food products. Starting at 0 until the end of the shopping trip.

Ratio

(continuous) Health label

Green Blue

The percentage of green and blue health labels in a specific product category at time t.

Ratio

(continuous) NonFood product

(Lagged, t-1)

A lagged dummy indicating 1 if the previous product bought was a non food product .

Ratio

(continuous) Table 3: overview of the main variables in model 1

Model 2 Variable name Variable explanation Scale

Dependent variable

Healthiness end of trip basket

A variable indicating the healthiness of the end of trip basket of a consumer

Ratio (continuous) Independent variables Total amount of products bought (self-control)

The total amount of food products bought by a consumer during a shopping trip.

Ratio

(continuous) Health label

Green/Blue

The weighted average of green and blue health labels.

Ratio

(continuous) Percentage Non

Food products in the end of trip basket

A variable indicating the percentage of non food products bought by a

consumer during a shopping trip

Ratio

(continuous)

(31)

31

3.3.7 Control variables

Next to the key variables several control variables were included in both models. These were included in order to test for their potential effects. Moreover, these variables were included in both models because they might explain variation which otherwise would have been caught by the independent variables in the models (Leeflang et al. 2015). A total of 4 control variables was included in both models. First, the variable 'week' was included. This categorical variable indicates in which week (week 1 until 6) the shopping trip took place. This variable is perceived interesting since consumers are expected to have strong health intentions in week 1. Moreover, retailer Plus performed an health intervention in week 2 and week 3. Second, the variable 'store' was included which indicated in which store the shopping trip took place. Although the stores are similar in size, they differ in type of neighborhood. Therefore, this variable is perceived interesting. Third, the variable 'time of the day' was included. This categorical variable indicates at what part of the day the shopping trip took place. It contains four categories: morning (from 09:00-12:00), early afternoon (12:01-15:00), late afternoon (15:01-18:00) and evening (18:01-21:00). This variable is perceived interesting in order to investigate if customers show different shopping behavior at different points in time. Lastly, the 'mean shopping duration per product' was included in both models. This variable was calculated by dividing the shopping time of a shopping trip by the amount of products bought during that shopping trip. This variable is perceived interesting in order to investigate if customers that think longer about a product decision show different shopping behavior.

3.4 Model proposition

Two models were estimated for this research. For the first model, the data contains multiple observations per customer. Therefore, a linear model would not be appropriate because this would violate the independence assumption (Leeflang et al. 2015). Hence, multiple

(32)

(Torres-32

Reyna 2007). In order to determine which model to use, one should therefore focus on the purpose of the research. This paper focuses on making inferences about healthy shopping behavior and thus focuses on individuals with certain shopping characteristics. Therefore, a random-effects model is more appropriate since this estimation procedure models individual differences by assuming different random intercepts for each customer. More specifically, in order to resolve the non-independence, this model assumes a random effect for each customer. Moreover, model 1 contains variables for which there is no within-group variation so a fixed-effects model would not have been appropriate since this type of model does not include these variables (Torres-Reyna 2007).

For the second model a ordinary least square regression was estimated with the healthiness of the end of trip basket as dependent variable.

3.5 Model specification 3.5.1 Model 1

Where:

 i= Unique ID for each customer: 1,2,3,4,...,I

 j= product category j: 1,2,3,4,...,J

 t= Nth purchase: 1,2,3,4,..., T

 = the relative healthiness of a single product choice in category j made by individual i at time t.

 = constant

 = parameter estimates

 ₌the cumulative sum of the healthiness since a positive inflection point for customer i at t-1.

= the cumulative sum of the unhealthiness since a negative inflection point for customer i at t-1.

(33)

33

= the percentage of green health labels in the product category j at time t. =a dummy indicating if individual i purchased a non food product at t-1.

= the number of purchases individual i did including non food purchases at time t. = a moderating effect of the cumulative sum of the healthiness and the number of purchases for individual i at time t-1.

= a moderating effect of the cumulative sum of the unhealthiness and the number of purchases for individual i at time t-1.

= a moderating effect of the percentage blue health labels in the product category j and the number of purchases for individual i at time t.

= a moderating effect of the percentage green health labels in the product category j and the number of purchases for individual i at time t.

 = time invariant error component

 = a vector of control variables: * Week

- Week 1 (Base category) - Week 2 - Week 3 - Week 4 - Week 5 - Week 6 * ShopID

- Shop 1(Base category) - Shop 2

- Shop 3

* Time of the day

- Morning (09:00-12:00) (Base category) - Early afternoon (12:01-15:00)

- Later afternoon (15:01-18:00) - Evening (18:01-21:00)

(34)

34

3.5.2 Model 2

Where:

 = the healthiness of the end basket of customer i at time t.

 i = Unique ID for each customer: 1,2,3,4,...,I

 t = time t

 = constant

 β1,2,3,...8 = parameter estimates

 =total amount of food purchases customer i did during the shopping trip at time t. =percentage of non food products that the basket of customer i contains at time t. =weighted average of blue health labels in the shopping basket of customer i at time t.

= weighted average of green health labels in the shopping basket of customer i at

time t.

 = a vector of control variables * Week

- Week 1 (Base category) - Week 2 - Week 3 - Week 4 - Week 5 - Week 6 * ShopID

- Shop 1 (Base category) - Shop 2

- Shop 3

* Time of the day

- Morning (09:00-12:00) (Base category) - Early afternoon (12:01-15:00)

- Later afternoon (15:01-18:00) - Evening (18:01-21:00)

(35)

35 -1 ,5 0 0 -1 ,0 0 0 -5 0 0 0 5 0 0 H e a lt h in e ss

4 Preparation of the data

This chapter starts with some preliminary plots and analyses of the key variables to explore outliers, oddities and variation over time. First, a boxplot was made of the variable

healthiness. Next, the variable healthiness was plotted over time. Lastly, the healthiness and the cumulative healthiness throughout the shopping trip were visualized.

4.1 Healthiness variable

Figure 5 displays a Boxplot of the dependent variable: healthiness. The Boxplot indicates three outliers with a healthiness ranging from -1000 until -1500. As discussed in section 3.3.1, the healthiness of a single product is calculated by distracting the amount of calories (per 100 gram) of a single product from the baseline of the corresponding product category. Hence, these three outliers seem like an oddity and are therefore further investigated.

(36)

36 -6 0 0 -4 0 0 -2 0 0 0 2 0 0 4 0 0 H e a lt h in e ss

actually contain 450 calories per 100 gram. Again, the new amount of calories (450) was added and the baseline for this category was recalculated.

Product description Amount of calories Product Category

Baseline Healthiness New amount calories New baseline New healthiness Party croissants 1440 Bread 283,91 -1156,09 345 254,08 -53,53 Party croissants 1440 Cheese 318,20 -1121,80 345 254,08 -53,53 Biscuit mini's 2075 Chocolate 524,88 -1550,12 450 519,20 69,20

Table 5: outliers of dependent variable model 1: healthiness

Figure 6 displays the new Boxplot (after changing the oddities). This Boxplot shows that there are no striking outliers anymore.

4.2 Healthiness over time

4.2.1 Healthiness per week

The data contains the first six weeks of 2016. Please recall from section 3.1 that there are several interesting events in the data. In week 1: consumers are expected to have strong health intentions due to 'the good intentions effect' and in week 2 and 3 there was a health

intervention by Plus (variation=benefits weeks). Figure 7 displays the 'mean healthiness' per week. More specifically, the healthiness (baseline-kc per 100 gram) for all single product choices in week 1 were summed and that number was then divided by the amount of products bought in week 1. Therefore, week 1 is the mean of the aggregated healthiness for that week. The figure shows that the mean healthiness is highest for the first two weeks of January. This could be due to the 'good intentions effect' as discussed in section 3.1, where customers start

(37)

37

the new year with strong health intentions. However, this good intentions effect is expected to be strongest in week 1 and after that slowly level off each week. Therefore, the fact that the mean healthiness is even higher in week 2 compared to week 1 could be due to the health intervention of Plus in that week (variation=benefit weeks). However, this health intervention was also taking place in week 3 where the mean healthiness is lower, but still higher than week 4,5 and 6. Interesting is the low healthiness in week 4 which could be a post-promotion dip after the health intervention in week 2 and 3. To conclude, Figure 7 shows some

preliminary evidence of a 'good intentions' effect in week 1 and a positive effect of the health intervention: variation=benefit weeks in week 2 and 3. Further analyses will need to clarify this.

Figure 7: mean healthiness per week 4.2.2 Healthiness per date

Figure 8 is a little more in-depth and displays the 'mean healthiness' per date. The calculation method is similar to the calculation method of the mean healthiness per week, but now per date. Week 1 (3-9 January) overall seems to be quite healthy. More interesting is the peak in the mean healthiness on the 10th of January, since this the start date of the health intervention by Plus (variation=benefit weeks). Also, 11 and 12 January seem to show healthy peaks compared to the other dates, which could also be due to this health intervention. Week 3 (17-23 January) starts less healthy, however at the end of week 3 the healthiness increases which could be due to the fact that these are the last days of the health intervention (ending at 23th of January). Week 4 (24-30 January), Week 5 (31 January - 6 February) and Week 6 (7-13 February) seem to be less healthy compared to week 1 and 2. Further analyses will need to clarify this. 0 2 4 6 8 1 2 3 4 5 6 M e an H e al th in e ss

(38)

38

4.3 Healthiness throughout the shopping trip

In order to gain more insights in the healthy shopping dynamics, the development of the healthiness throughout the shopping trip is visualized. Customers on average made 20 food purchase decisions with a standard deviation of 10.85. Therefore, it was decided to visualize the first 25 food purchases. Figure 9 shows the healthy shopping dynamics in a consecutive purchase process. The calculation method is similar to the calculation method of the mean healthiness per week but now per purchase. Therefore, food purchase 1 is the mean aggregated healthiness of this product choice for every consumer. Please note that the healthiness of each product is corrected for the baseline of the corresponding product category, therefore it does not matter in which sequence customers visited the different product categories. The trend line shows a fluctuating pattern throughout the shopping trip. Hence, customers start their shopping trip healthy (although the healthiness is slowly declining at each purchase) and after 13 food purchases a negative inflection point takes place, leading to a set of unhealthy purchases. This could be some preliminary evidence of a lack of self-control after so many purchases. Then at food purchase 19 a positive inflection point takes place, leading to a set of healthy product choices again. The mean healthiness is the lowest at food purchase 24. As the figure shows, customers end their shopping trip with unhealthy product choices. This could be due to the fact that they lost self-control when heading to the cash register.

CONSTRAINING THE OBESITY EPIDEMIC BY UNDERSTANDING THE ROLE OF HEALTHY SHOPPING DYNAMICS

CONSTRAINING THE OBESITY

EPIDEMIC BY UNDERSTANDING THE

ROLE OF HEALTHY SHOPPING

DYNAMICS

by Sanne Schreurs

Master Thesis

CONSTRAINING THE OBESITY EPIDEMIC BY

UNDERSTANDING THE ROLE OF HEALTHY

SHOPPING DYNAMICS

Management summary

Preface

Table of contents

1 Introduction

2 Theoretical framework

3 Methodology

(x

)

(

)

(

-x

)

4 Preparation of the data