TIME FOR VICE?

i

TIME FOR VICE?

DYNAMICS IN CONSUMERS’ ONLINE GROCERY SHOPPING BEHAVIOUR

ii

TIME FOR VICE?

DYNAMICS IN CONSUMERS’ ONLINE GROCERY SHOPPING BEHAVIOUR

Master thesis

MSc Marketing Management & MSc Marketing Intelligence Faculty of Economic and Business

University of Groningen

Author:

Rosanne Heijligers

Date:

June 26, 2017

Address:

Donkerstraat 17C

3511 KB Utrecht

Phone number:

+31(0)614944006

Email address:

r.a.h.heijligers@student.rug.nl

Student number:

2799154

Department:

Faculty of Economic and Business

First supervisor:

dr. H. Risselada

iii

MANAGEMENT SUMMARY

Despite the fact that consumers are concerned about their health, obesity statistics still illustrate that consumers experience difficulties with choosing healthy alternatives (virtues) rather than unhealthy products (vices). Given this growing interest in healthy shopping dynamics, and next to the development of new online channel grocery alternatives, a better understanding of the role of the online grocery retail channel and vice consumption is needed. As an initial step in addressing this need, this study elaborates on the few studies done in this field (Huyghe, Verstraeten, Geuens, & Van Kerckhove, 2017; Milkman et al., 2010) and adds novel time effects to it. To this end, this study aims to fill an important gap by shedding light on the role of time in stimulating vice purchases. The objective of this study is to find out to what extent do (1) Online Lead Time, (2) Online Grocery Shopping Maturity, (2) Time of the Day, (4) Time before Meal (‘hunger’), and (5) Fresh-Start-Effects, influence the choice of vices in an online grocery channel. In this research, theories from self-control (Baumeister, 2002), visceral states (Loewenstein, 1996), construal level (Trope & Liberman, 2007), hot-cold empathy gap (Loewenstein, 1998), vividness (Loewenstein, 1996), and Fresh-Start-Effects (Dai, Milkman & Riss, 2014) are integrated to provide a better understanding of how different time characteristics affect consumers’ online vice consumption. By using a rich panel dataset of more than 59120 Dutch online grocery shoppers over 2 years, a linear regression model with consumer fixed effect was estimated to measure the impact of time on consumers’ Share Of Vice (SOV). The results show that different time effects significantly influence consumers’ vice consumption. The results show that as the delay between order completion and delivery increases, consumers order a relatively lower amount of vices, which is line with construal level theory and consumers’ biases toward the present. As consumers become more experienced online, they order relatively less vices. The effect of Online Lead Time on consumers’ SOV is stronger compared to the effect of Online Maturity. In support with self-control depletion theories (Baumeister, 2002), this study found that the same consumers order relatively more vices during late evening and before meals moments (lunch and dinner). In terms of Fresh-Start-Effects (Dai et al., 2014), the results confirm that during the first month of the year, at the first day of the week, at the beginning of the month and after a public holiday consumers order a relatively lower amount of vices online. Given the managerial importance of vice promotions on SOV, the effect of vice promotions is taken into account. The results show that as the number of vice promotions increases, consumers’ SOV increases. Contrary to expectations, this effect is positively moderated by the month January (Fresh-Start-Effect), which suggests that in the month January consumers are especially appealed by vice promotions. The findings of this study suggest that time effects have significant implications for consumers, grocery retailers, and policy makers. Finally, directions for future research are discussed.

Keywords: Unhealthy food choices, Time Effects, Self-control, Online Lead Time, Online Buying Experience,

iv

PREFACE

A five month journey comes to an end with this final piece of work. The journey started of writing this piece of work started when I heard that I got the opportunity to write my thesis at one of the largest online grocery retailers in the Netherlands, which operates in a field that appeals to me: the dynamic world of online retailing. At the start of the thesis process, my external supervisor Roel and I discussed several research opportunities. It was challenging to come up with an original thesis subject, which was both managerially and academically relevant, and not studied before. By sparring with Roel and by reading a lot of papers from different research fields, I came up with the idea to research the relation of time and consumers’ unhealthy behavior.

During the past months I encountered several analytical challenges. I experienced a deep learning curve, and learned to work with different statistical tools and models. Next to that I also learned how an online retailer operates behind the scenes as well as in the field. All in all the process was challenging, but I am satisfied with the outcome and grateful for all the things that I have learned.

I am also grateful towards the focal online retailer for providing me this research opportunity. I would like to thank dr. Hans Risselada for guiding me through the thesis process, providing me with valuable and critical feedback and challenging me to continuously improve my thesis. In addition, I would like to thank my external supervisor Roel Willems for the great sparring sessions and for providing me useful feedback. Further, I would like to thank prof. dr. ir. Koert van Ittersum for taking time to read my work. Moreover, I would like to thank Jasper Hidding for his support and providing me feedback on my thesis. I would also like to thank my boyfriend Rutmer Faber, for sparring with me, listening to my thesis stories almost every day and being there for me. Finally, I want to like to thank my family, friends, and my colleagues for their support during the period of this research project.

v

TABLE OF CONTENTS

1. INTRODUCTION ... 1

2. THEORETICAL BACKGROUND ... 6

2.1 Online Grocery Shopping conceptualization ... 6

2.2 Vice and Virtue products ... 7

2.3 Drivers of Vice purchases ... 7

2.3.1 Self-control. ... 8

2.3.2 Proximity. ... 8

2.4 Time Effects influencing Vice purchases ... 9

2.4.1 Online Lead Time. ... 9

2.4.2 Online Grocery Shopping Maturity. ... 11

2.4.3 Time of the Day. ... 13

2.4.4 Time before Meal. ... 14

2.4.5 Fresh-Start-Effect. ... 15

2.6 Vice Promotions influencing Vice purchasing ... 16

2.8 Control Variables ... 17

2.8.1 Week numbers. ... 18

2.8.2 Day order delivered. ... 18

2.8.3 Days since last delivery. ... 18

2.8.4 Times opened from submitted. ... 18

2.8.5 Days between webvisit order. ... 18

2.8.6 Number of products and Monetary value promotional products. ... 18

2.8.7 Year effects ... 18

2.8.8 Seasonal Influences. ... 19

2.8.9 Promotional meal box. ... 19

2.8.10 Home Delivery and Pick Up Point. ... 19

vi

4.2.4 Multicollinearity. ... 37

4.4 Results ... 39

4.4.1 Parameter Estimates... 39

4.4.2 Online Lead Time. ... 40

4.4.3 Online Grocery Shopping Maturity. ... 41

4.4.4 Time of the Day ... 41

4.4.5 Time before Meal. ... 41

4.4.6 Fresh-Start-Effects. ... 41

4.4.7 Vice Promotions and Interactions. ... 42

4.4.8 Control variables. ... 42

5. CONCLUSION ... 45

5.1 Discussion ... 46

5.1.1 Time Effects. ... 46

5.1.2 Vice promotion and interaction Vice promotion * First Month. ... 48

5.2 Consumer, Public Policy, and Managerial implications ... 49

5.3 Limitations and further research recommendations ... 51

LITERATURE ... 53

Appendix A – Variables Description ... 60

Appendix B - Panel effects ... 63

Appendix C – Pearson’s Correlation Matrix ... 64

Appendix D - Fixed effect model steps ... 65

Appendix E - Testing model assumptions ... 72

Appendix F – Multicollinearity ... 78

Appendix G - Coefficients alternative models ... 83

Appendix H - Robustness checks ... 85

LIST OF TABLES

Table 2 Summary of the hypotheses. ... 20

Table 3 Summary Statistics and Variable Descriptives... 24

Table 4 Overview and description of variables ... 60

Table 5 Testing for Panel effects ... 63

Table 6 F-test ... 63

Table 7 Hausman test ... 63

Table 8 Visualization of Pearson’s Correlation Matrix ... 64

Table 9 Model building steps ... 65

Table 10 Summary of model building steps ... 33

vii

Table 12 Testing for autocorrelation ... 72

Table 13 Testing for heteroskedicity ... 73

Table 14 Testing for non normality ... 74

Table 15 Bootstrapping estimates ... 76

Table 16 Variance Inflation Factors of initial model ... 78

Table 17 Multicollinearity check for four models ... 79

Table 18 VIF and tolerance(1/VIF) per model ... 81

Table 19 Goodness-of-fit measures and information criteria per model ... 82

Table 20 Variance Inflation Factors of the variables ... 38

Table 21 The Fixed effect model ... 39

Table 22 Random and Pooled OLS model ... 83

Table 23 Results Paired t-test... 41

Table 24 Robustness checks ... 85

Table 25 Overview of the hypotheses and findings ... 45

LIST OF FIGURES Figure 1 Conceptual Model ... 19

Figure 2 Difference between confirmation date and delivery date (Online Lead Time) ... 31

Figure 3 SOV and Order Lead Time ... 31

Figure 4 Order Time of the Day ... 32

Figure 5 Distribution of PUP and HD ... 32

Figure 6 Delivery and order day of the week ... 32

Figure 7 SOV during week days ... 32

Figure 8 SOV over months ... 32

Figure 9 Days between webvisit order ... 32

Figure 10 Autocorrelation Function (ACF) with pooled OLS and fixed effect estimation ... 35

Figure 12 Residuals plotted against log SOV and distribution of log SOV ... 37

Figure 11 Histogram and Normal Q-Q plot of residuals ... 74

1

1. INTRODUCTION

Many countries are facing an obesity epidemic. In the Netherlands for example, overweight rates are at an alarming rate: 43% of the Dutch population of 4 years or older is overweight (BMI ≥ 25 to <30), of which 12% is indicated to be obese (BMI ≥ 30) (CBS, 2016). Accordingly, one out of ten Dutch citizens is considered to have obesity. Currently, this obesity is one of the greatest public health challenges of the 21st century in modern Western society. Its occurrence has more than tripled in many European countries since the 1980’s, and the number of those affected continues to rise, particularly among children (World Health Organization, 2016). Obesity can result in critical consequences, such as cardiovascular diseases, musculoskeletal disorders, and different types of cancers (World Health Organization, 2016). The World Health Organization (2017) estimates that in many developed countries, obesity is responsible for 2-8% of all health cost. Obesity is fundamentally a consumption problem, resulting in a continued imbalance of energy intake in excess of expenditure, leading to storage in the form of fat and subsequent weight gain (Moore, Wilkie & Desrochers, 2016). Although obesity is a multifactorial disease, overconsumption of unhealthy energy-dense, and nutrient-poor foods with higher concentrations of fat, sugar, and salt is a main driver of the epidemic (Drewnowski, 2005). These vice products are appealing products that typically satisfy short-term benefits (e.g. tastiness) but have negative long-term outcomes (e.g. healthiness) (Wertenbroch, 1998).

To get a better understanding of shoppers’ vice consumption, several studies have focused on contextual cues that affect consumer’s choice. In fact, previous academic research has shown that ambient light (Biswas, Szocs, Chacko & Wansink, 2017), lateral display patterns (Romero & Biswas, 2016), point-of-sale nutrition scoring systems (Nikolava & Inman, 2015), available items (Sela, Berger & Liu, 2009), low-fat nutrition labels (Wansink & Chandon, 2006) and default options (Samuelson & Zeckhauser, 1988) exert significant influence on consumers’ decision to consume vice products.

Next to these so called choice architecture effects on vice consumption, time effects can also be associated with consumers’ consumption behaviour. For example, Dai, Milkman & Riis (2014) document the 'Fresh-Start-Effect’ theory, indicating that consumers are more likely to begin dieting and exercising at the start of a temporal cycle, such as the beginning of the year and after holidays. Other academic research has shown that people are more likely to consume indulgent vice food at the end of the day (Baumeister, 2002) and when consumers experience hunger (Read & van Leeuwen, 1998).

2

More specifically, predictions indicate that in the Netherlands one out of six Dutch citizens will shop their groceries online in 2017, resulting in an increase of online grocery revenue over 30% (Multiscope, 2016). This highlights the potential and the importance of a better understanding of this relatively new channel. Accordingly, despite this radical growth of the online grocery retail channel and the societal, political, and food industry’s urge to promote healthy food, there is still very limited research in this area. To the best of author’s knowledge, only two studies investigated the effect of (un)healthy choices in a real online grocery context. First - most recently - Huyghe, Verstraeten, Geuens & Van Kerckhove (2017) found that the multi-channel consumers purchase less vice products online compared to the offline channel. This channel difference in consumption behaviour was explained by the presentation mode of the vice product: physically in a physical store and symbolically online. The online symbolic presentation decreases the vividness of products, which consequently decreases consumers’ striving for instant gratification. In the end, this lead to the consumers purchasing less vice products. Second, a study by Milkman, Rogers & Bazerman (2010) found that consumers select healthier food online for the distant future compared to the near future. They showed that as the delay between order completion and delivery time increased, consumers purchase less and order a higher percentage of healthy products and a lower percentage of vices.

This research will elaborate on these latter two studies by responding on the specific additional research call of Huyghe et al. (2017), and by testing the generalization of the results of Milkman et al. (2010) study in a different country and time context. The study of Huyghe et., al (2016) was only focused on the difference between online and offline consumption for a relatively short period, in which they assessed that online and offline shoppers made a distinct choice at one specific moment in time. However, familiarity of constitutive elements (i.e. online grocery website, products) determines the vividness of a mental thought (D’Argembeau & Van der Linden, 2012). In this research, it is proposed that this familiarity with the online channel and its products increases over time because of online grocery maturity (experience), and this so may increase vividness over time which consequently may affect consumers’ online grocery behaviour. Huyghe et al. (2017) call for additional research on this long-term familiarity effect of changes in online grocery shopping behaviour.

3

consumers who order for the near future, because consumers “should” self-execute more influence over choices in the further future (Rogers & Bazerman, 2008).

Building on these latter two papers, this research adds novel time effects in the vice consumption context. Inspired by Milkman et al. (2010), this study will elaborate on additional time effects which can influence consumers’ vice behavior, next to the one time effect (Online Lead Time) proposed by Milkman et al. (2010). In this research several time effects are proposed which can explain consumers’ vice online consumption. By this, a more comprehensive understanding of vice consumption in the context of time can be provided. This is done by building on self-related control theories in the domain of time (Dai et al., 2014; Baumeister, 2002; Read & van Leeuwen, 1998). In this research it is postulated that the “Fresh-Start-Effect” affects consumers’ online proportion of vices. Additionally, it is proposed that when the same consumers shop at the end of the day or before a dinner moment - in a state of hunger – they order relatively more vices.

1.1 Aim of research

The objective of this study is to identify how different time effects affect consumers’ choices for vices in an online channel. More specifically, this research seeks to explore the effect of Online Grocery Shopping Maturity, Online Lead Time, Time before Meal, Time of the Day, and Fresh-Start-Effects on online shoppers’ behaviour and specifically on consumers’ choice for vices products. Given this objective, the following research question is addressed:

To what extent do (1) Online Lead Time, (2) Online Grocery Shopping Maturity, (2) Time of the Day, (4) Time before Meal (‘hunger’), and (5) Fresh-Start-Effects, influence the choice of vices in an online grocery channel?

An extensive panel dataset of a Dutch online grocery retailer is used. This dataset is comprised of consumers’ online shopping behaviour over time in the period January 2015 - April 2017.

1.2 Relevance

This research contributes to both marketing research and practice. The contributions to academic research are the following:

4

component, this research further refines and extends previous research on online buying experience effects (Campo & Breugelmans, 2015; Kim et al., 2008), by examining the effect of online grocery shopping maturity on consumers’ choice of vices online.

2) This study replicates the findings of Milkman et al. (2010) in a different continent and at a different moment in time. Therefore this study can contribute to the generalizability of the findings from Milkman et al. (2010).

3) This study contributes to self-regulation literature by including the Time of the Day as a potential influencer of self-control strengths. The Time of the Day effect has been linked to unplanned grocery spending in a physical store (Kollat & Willet, 1967), but not yet in an online grocery store. 4) This study includes a potential new construct which may be influencing consumers’ online

purchase behaviour: consumers’ present hunger state. By incorporating this, the current research extends the hot-cold empathy framework (Loewenstein, et al. 1998) to an online context. So far, no empirical research has tested this hunger-state and its effect on consumers’ behaviour in an online grocery environment.

5) This study incorporates the effect of temporal landmarks on consumers behaviour. In fact, this study adds to the applicability of the ‘Fresh-Start-Effects’ (Dai, et al., 2014) in a online grocery context. This effect has been found in an online ordering restaurant setting (Van Epps, Downs & Loewenstein, 2016), but not yet in an online grocery retail setting.

6) Lastly, this research contributes to the growing number of online store papers. Previous online store studies mainly focused on durable goods, these settings are fundamentally different from grocery settings (e.g. Nepomuceno, et al., 2014). Hence, implications differ substantially.

5

product assortment and promotional strategy. Moreover, insights in the ‘Fresh-Start-Effects’ online could improve online promotional strategy. For example, an online grocery retailer could adjust companies’ promotional strategy if consumers are more inclined to purchase healthy food after a summer holiday. All in all, this research provides insights for retailers, public policy makers, and consumers.

1.3 Outline

6

2. THEORETICAL BACKGROUND

This chapter provides an overview of relevant prior academic research related to online grocery shopping and vice products. First, the concept of online grocery shopping will be discussed and the key differences between online grocery shopping and other forms of online shopping will be explained. The concept of vice and virtue products will be discussed next. After this, the main drivers of vice purchases and related dominant theories will be discussed.

After this general theoretical overview, five different time effects are proposed which may influence the amount of money spent on vices number relative to the total amount of money spent online (Share Of Vice). In this second part, different hypotheses are proposed which may influence consumers’ Share Of Vice products (SOV) online All hypotheses are visually summarized in a conceptual model, which can been found at the end of this chapter.

2.1 Online Grocery Shopping conceptualization

In this research online grocery is the defined as “the process of purchasing groceries from home in

an electronic way and either let the groceries delivered at one’s house or collecting them at a store or at a pick-up point” (Cagliano, De Marco & Rafele, 2017, p.3). Online grocery has several benefits over

7

and consumers are not prepared to spend much time and effort to search for the ‘optimal’ product (Hoyer & MacInnis, 2010).

2.2 Vice and Virtue products

Within this research, unhealthy and healthy products are referred to as vices and virtues, in line with previous used terminology in other research on unhealthy (vice) and healthy (virtue) foods (Huyghe et al, 2017; Liu et al, 2015; van Doorn & Verhoef, 2011). The differences between these two concepts are based on the timing of their payoff. Vice products (e.g. chocolate bar) lead to immediate pleasure, but they also provide negative outcomes in the long run (e.g. gaining weight). Contrary, virtue products (e.g. apple) are less appealing at the point of decision, but show less negative outcomes in the long run and are therefore regarded as a well-considered choice. Virtue products thus provide more utility in the long run compared to vice products but less utility in the period shortly after it is perceived (Read, Loewenstein & Kalyanaraman, 1999; Wertenbronch, 1998). Researchers have used similar but different terms to categorize the extent to which products satisfy short-term versus long-term benefits. Examples are ‘wants’ and ‘should’ products (Milkman et al, 2010), and ‘hedonic’ and ‘utilitarian’ products (O'curry & Strahilevitz 2001). Related to these vice products are impulse purchases, which are often associated with negative consequences for the consumer (Rook, 1987).

2.3 Drivers of Vice purchases

Many of the choices that get us into trouble involve simple choices between virtues and vices. As such, consumers face choices when deciding between a virtue option (e.g. green snacks) that is in line with their long-term desire to live a long and healthy live and a vice option (e.g. chocolate bars) that is easier to give an immediate gratification. This type of intrapersonal conflict is in literature referred to as the ‘multiple selves phenomenon’ (Ainslie & Haslam, 1992): people have a feeling of two selves inside them, one is more focused on the present and the other one in more future oriented – both battle for control of the behaviour (Loewenstein, 1996). These conflicting drivers thus create a self-control dilemma, in which qualitatively different kind of emotions – hedonic and self-conscious – opposed each other in valence (Giner-Sorolla, 2001). More specifically, research in this field found that choice exposure may activate two types of processes: one affective process and one cognitive process (Shiv & Fedorikhin, 1999). Affective processes occur automatically, whereas cognitive processes occur in a more controlled situation. The latter one results in more cognition about the consequences of the choice option. If the availability of processing (self-control) resources is temporarily depleted, the choice decision is often made automatically (Shiv & Fedorikhin, 1999).

8

temper these affective response (self-control). Next to exerting self-control, consumers’ unplanned behaviour can be influenced by proximity (Vohs & Faber, 2007). At sufficient levels of intensity, these elements cause consumers to behave contrary to their long-term self-interest and let them chose vice products (Loewenstein, 1996). Both these two drivers of vice consumptions will be explained up next.

2.3.1 Self-control. When a consumer is deciding to purchase a healthy product, a certain level of

self-regulation is required. Here, the terms ‘self-control’ and ‘self-self-regulation’ are used interchangeably, and

both refer to the same thing: “the self's capacity to alter its own states and responses” (Baumeister, 2002, p.670). Self-control behaviours are generated to increase the long-term benefits of the individual.

According to Baumeister (2007) sufficient self-control depends on four elements: standards, a monitoring process, self-regulatory strength and motivation. ‘Standards’ are related to goals, norms, ideals and other guidelines that state the desired response. When a consumer knows exactly what he wants, he is less likely compared to others to indulge in unplanned buying. However, conflicting goals can weaken self-control and make consumers more vulnerable. Second, ‘monitoring’ is related to keeping track of important behaviour. Third, the process required to change the self are often difficult and ask for some will power: ‘self-regulatory strength’. The fourth element required for self-control is a person’s ‘motivation’ to realize his goal and ideals. Still if a person has clearly defined standards, monitoring is successful, and the person has rich resources, the person might still fail to self-regulate because he is not motivated to realize the goal.

The amount of self-control one consumes is a limited resource. Exercising self-control requires some self-control strength, which consequently reduces the amount of strength available for succeeding self-control efforts. Maruven & Baumeister (2000) point to three mechanisms that likely cause subsequent self-control effort to fail: dealing with stress, regulating negative affect and resisting temptations. Accordingly, this limited resource model forecasts that when a person attempts to engage in various demanding self-regulatory task repeatedly, the change of success at resisting all of them is reduced. Hence, after executing several self-control attempts a person’s resource become temporarily depleted. This temporarily depletion of the self-control makes a person more vulnerable and more likely to purchase impulse temptations (Vohs, Baumester & Ciarocco, 2005).

Fortunately, self-resource depletion can be overcome. Self-control resembles a muscle, which can be trained over time in order to make in stronger and increases its power to self-regulate (Muraven, Baumeister & Tice, 1999).

9

affective visceral desires will overtake the cognitive systems (Loewenstein, 1996). These visceral states includes the influence of emotions, drives and somatic sensations such as hunger and anger (Loewenstein, Prelec & Shatto, 1998).

Sensory distance decreases as the degree of firsthand contact with a product increases. When a product is physically absent, and only its verbal brand name is presented, then the sensory distance is high. Consumers are then thinking about the product in abstract terms (Kardes, Cronley & Kim, 2006). For example, the word “chocolate” does not activate the same affective, cognitive and behavioural responses that are provoked by a real chocolate bar. Hence, it is easier for a consumer to decide not to eat the tempting chocolate bar when exposed to the word only (Shiv & Fedorikhin, 1999). Comparable to physically absent products, symbolic product presentations (e.g. pictures) are lower in sensory proximity, and their vividness decreases compared to product presented physically (Kardes, et al, 2006). This symbolic presentation was found to decrease products’ vividness and diminishes consumers’ striving for immediate gratification, and consequently lowers the tendency to purchase vice products (Hughe, et al. 2017). Next to sensory and physical proximity, temporal proximity can also influence consumers’ behaviour towards vice products. Consumers for example have the propensity to change their food decision in the direction of vice as the moment of consuming approaches (Read, Loewenstein & Kalyanaraman, 1999). Temporal proximity will be explained further in 2.4.1.

2.4 Time Effects influencing Vice purchases

In this research, five different time effects are investigated whom may influence consumers’ SOV. Therefore, different hypotheses are proposed (Table 2), which are all visually summarized in the conceptual model (Figure 1).

2.4.1 Online Lead Time. When consumers purchase their groceries online, consumers make choices about

their future consumption. Online delivery time imposes a temporal time delay, such that consumers cannot immediately consume the product they ordered. As such, temporal proximity can differ per consumer: some consumers choose to deliver their groceries the next day while others prefer their groceries to be delivered at the end of the week. In this section the potential effects of the Online Lead Time on SOV will be discussed by elaborating on three different time theories: (1) present bias, (2) hot and cold states, (3) and Construal Level theory.

Present bias. Prior research has shown that consumers are likely to choose a virtue product over a vice

10

a time discounting ‘hyperbolic function’: as the temporal distance of an outcome increases, the decline in the perceptual value of the outcome is initially steep and then becomes moderate (Trope & Liberman, 2007). This leads consumers to overvalue the present utility relative to the future utility, and thus prefer vice choices over virtuous choices. When decisions are framed to postponed consumption, consumers are likely to show the desires of the ‘should’ self, relative to the decisions to the close future (Rogers & Bazerman, 2008). This indicates that time postponements not only discriminate between ‘now’ and “later”, but also (to a certain degree) between ‘soon’ and ‘later’.

Hot and cold states. Research found that sensory cues and other external stimuli can be processed by two

different systems: ‘cool’ and ‘hot’ (Metcalfe & Mischel, 1999). The cool system is cognitive, reflective and responsible for self-control, while the hot system is emotional, reflexive and mainly driven by automatic responses to external stimuli. The researchers proposed that the effect of temporal distance depends on whether the outcomes are affected-based (hot) value or cognitive-based (cool) value. Metcalfe & Mischel (1999) showed that as temporal distance increases, the weight of cognitive outcomes increases and the weight of affective outcomes decrease in value. For example, as time horizon increases the value of a product is more likely to be on its healthiness (cool) than on its tastiness (hot). These researchers accordingly found that choices in the present tend to be visceral, while choices for the future tend to be virtuous (Metcalfe & Mischel, 1999).

Construal Level theory. Construal Level theory (CLT) proposes that the construal level - the abstractness of

the mental representation is formed - increases with temporal, spatial or sensory distance (Trope & Liberman, 2007). Higher abstraction levels lead to general mental representations that lack contextual detail, whereas lower abstraction levels lead to concrete, vivid, and specific mental representations that include rich contextual detail (Table 1). Hence, when consumers order their grocery further in advance, they are likely to mentally represent the groceries in terms of relatively broad classes, such as healthy food, and the types of choices are likely to be based on higher-order goals, such as eating healthy (Trope & Liberman, 2007). From a distant time view, consumers see the whole picture, whereas in a close time view, consumers tend to see the detail (Trope &

Liberman, 2007). In contrast, when consumers make a choice for the present they are more worried about the subjective experience, thereby focusing on low-level details, such as the tastiness of the product. As the reward comes (sensory, temporal, physical) closer, the probability that consumers’ affective visceral desires will overrule the cognitive systems increases (Loewenstein, 1996).

11

A lower abstraction level is related to vividness and the ease of which a past instance of an outcome can be remembered, producing a subjective probability via the ‘availability heuristics’ (Loewenstein, 1996). For example, the more vivid a product picture is, and the greater detail which they are recalled, the greater the emotional response (Miller et al., 1987). A real presentation of a product in a physical store enhances vividness, and makes it easier to sense the gratification arising from consuming the alternative compared to a symbolic presentation (Shiv & Fedorikhin, 1999). Thus, a more vivid product presentation increases the imagination of consuming the product, which makes these consumers more vulnerable to choosing vices.

These latter three theories forecast that food decisions made for closer future (i.e. tomorrow) will include an increased spending in vice products, whereas spending for the more distant future (i.e. next week) will lead to a decreased spending in vices. Applying this logic to the online grocery context, delivery times may impose a temporal time distance in that a consumer cannot immediate consume the product because the product is not immediately delivered. Online Lead Time is defined as the time between order confirmation and delivery (Milkman, et al. 2010).

Hence - in line with CLT - products in an online store have a higher temporal proximity, which subsequently decreases the vividness of the product compared to a physical grocery shop. The time discounting theory can also be applied more specifically to online grocery shops order lead time. In particular: do consumers make different purchase decisions online as the prospective duration (e.g. time between ordering and delivery) increases? To research this, the following hypothesis is proposed:

H1: Online Lead Time has a negative effect on the SOV in consumers’ online grocery basket.

2.4.2 Online Grocery Shopping Maturity. Consumers might adjust their purchase behaviours as they gain

12

Vividness over time. One variable that arouses visceral factors is products’ vividness (Loewenstein, 1996).

A more vivid product presentation increases the imagination of consuming the product, which makes a consumer more vulnerable to choosing vices (Huyghe et al., 2017). In this research, vividness is defined as information that is, “as likely to attract and hold our attention and to excite the imagination to the extent

that it is (1) emotionally interesting, (2) concrete and imagery-provoking, and (3) proximate in a sensory, temporal, or spatial way" (Nisbett & Ross, 1980, p.45). Vividness is often treated as a communication

attribute (Taylor & Thompson, 1982). For example, pictures and video type of information are assumed to be more vivid than information that lacks such attention receiving aids. Nisbett & Ross (1980) proposed different mechanisms that might produce this vividness effect, such as an increased accessibility of vivid information in memory and greater rehearsal of vivid material.

Sherer & Rogers (1984) suggest that the dominance of vivid information should increase with passage over calendar time, because of its greater mental availability. More specifically, these researchers tested the effect of vivid information in fear persuasive communications on attitude change. Despite little overall support for the increased vividness of persuasive communications with the passage over time (between 48hours time intervals), they found that the impact of information with high emotional interest and near temporal proximity was improved over time.

Moreover, events in familiar locations are more vivid than events in unfamiliar locations (Arnold, McDermott, & Szpunar, 2011). These researchers found that both location familiarity and temporal distance influence the vividness of location (Arnold, McDermott, & Szpunar, 2011). Further exploring the effect of familiarity, D’Argembeau & Van der Linden (2012) found that familiarity of constitutive elements (i.e. the envisioned location, persons and objects) largely determines the vividness of an episodic future thought. Additionally, they found that the effect of temporal distance on vividness was mediated by location familiarity. The benefit of increased location familiarity is that this could carry over to remembered details of that context (Robin & Moscovitch, 2014).

13

vivid context makes indulgent products more concrete, and so consumers are more vulnerable for cravings online (Loewenstein, 1996). Following this line of reasoning, the following hypothesis is proposed:

H2: Online Grocery Shopping Maturity has a positive effect on the SOV in consumers’ online grocery basket.

Arnold, McDermott, & Szpunar (2011) found that although both context familiarity and temporal distance influence the vividness of location, the effect was stronger for context familiarity compared to temporal distance. Applying this theory on the online grocery context, one can argue that Online Grocery Shopping Maturity (‘context familiarity’) may have a higher impact on vividness of the online channel, and thus to vices purchased online, compared to online lead times (‘temporal distance’):

H3: Online Grocery Shopping Maturity has a higher impact on SOV in online consumers’ grocery basket online compared to Online Lead Time.

2.4.3 Time of the Day. The amount of self-control one exerts is limited (Vohs & Faber, 2007). Exercising

self-control requires self-control strength, which consequently reduces the amount of strength available for succeeding self-control efforts (Vohs & Faber, 2007). This is especially interesting in consumers’ behaviour, as consumers make around 35,000 choices daily (Hoomans, 2015), of which 200 daily food decisions (Wansink & Sobel, 2007), while being exposed to around 5000 advertisements per day (Johnson, 2006).

Research found that breakdowns in self-control are more likely to occur in the evening than during the day (Baumeister et al, 1994). Similarly, diets are often broken late in the evening and people smoke and drink intensely at the end of the day (Baumeister & Heatherton, 1996). Hence, the daily cycle of life can be interesting for consumers’ control depletion (Baumeister, 2002). During sleep self-resources are restored, and these self-self-resources become gradually depleted during the day. Therefore, breakdowns in self-control are rare in the morning and are more gradually more likely (Baumeister, 2002). For example, people are less likely to break diets as the first thing in the morning. Baumeister (2002) suggests that indulgent buying and increased spending becomes more and more likely as the day continues. Kollat & Willet (1967) found support for this self-depletion effect at the end of the day. These researchers tested whether the number of unplanned purchases increases in a physical store as a function of time. These researchers found that Time of the Day was related to consumer difference in unplanned purchasing, because the number of different products purchased increased.

14

the person’s self-control while behaviour may become more impulsive. Following the above arguments, the following hypothesis is proposed:

H4: The end of the day (Late Evening) is positively related to the SOV in consumers’ online grocery basket, compared to the rest of the day.

2.4.4 Time before Meal. Visceral factors, such as hunger can influence consumers’ behaviour

(Loewenstein, 1996). Hungry people order more food than they can eat (the ‘eyes bigger than your stomach’ effect), prefer candy over fruit (Read & van Leeuwen, 1998) and demonstrate less self-control (Kirk & Logue, 1997). Prior research shows that hungry shoppers not only have higher grocery expenditure, they also tend to purchase more unplanned and higher-calories items (Gilbert, Gill & Watson, 2002; Tal & Wansink, 2013).

Temporal proximity powerfully influences visceral factors, such as hunger (Loewenstein, 1996). At low levels of hunger, people typically feel they can integrate visceral influence in their behaviour in a rational fashion. At moderate levels, visceral factors tend to create conflicts between the desired and actual behaviour, the ‘multiple selves phenomenon’. At higher levels, visceral factors take complete control of behaviour (Loewenstein, Prelec & Shatto, 1998). In an online grocery context temporal distance is present, as the products cannot be consumed immediately after purchase. As visceral impulses (i.e. hunger) demand immediate gratification (Marnet, 2008), temporal distance may reduce the tendency to satisfy current visceral states and thus lowers the tendency to purchase vices online.

15

Taken together, on the one hand temporal distances may impose a barrier to immediate gratification of visceral states and thus lowers the tendency to purchase vices online. However, consumers often engage in a hot-cold empathy gap in which current states determines future choices. Interestingly, present states often effect future choices unconsciously. So far, no empirical research has tested this hot-to-cold effect in an online grocery environment. To test this effect the following hypothesis is proposed:

H5: Time before Meal (‘hunger state’) has a positive effect on the SOV in consumers’ online grocery basket.

2.4.5 Fresh-Start-Effect. The beginning of a new year is commonly known as the time of the year when

thousands of people commit themselves with unusual energy to achieve personal set goals, such as eating more healthy (Norcross et al. 2002). These type of moments are called temporal landmarks, which is referred to as distinct events that are different to the everyday routine of life (Peetz & Wilson, 2013). This could be public events (e.g. a national election, holidays, new school year), personal-relevant events (e.g. becoming parent, birthday, divorce), or calendar reference points (e.g. Christmas, beginning of new weeks, years).

Prior research has shown these temporal landmarks determine the passage of time and create discontinuities in our memories, experiences and time perceptions (Robinson, 1986). These temporal landmarks separate consumers’ life in distinct mental accounting periods (Peets & Wilson, 2013). Dai et al. (2014) called this the Fresh-Start-Effect, indicating that consumers are more likely to begin dieting and exercising at the start of a new cycle. Two psychological mechanisms were proposed that explained this Fresh-Start-Effect (Dai et al., 2014). First, new mental accounting periods such as temporal landmarks separate the current self from past imperfections, driving consumers to behave in line with their new, positive self-image and thus motivates aspirational behaviour (Dai et al., 2014). Second, temporal landmarks induce people to take a big-picture of their lives, and thus focus more on achieving consumers’ goals (Dai et al., 2014). Temporal landmarks can therefore serve as a disruption to decision-making processes and direct motivation to aspirational behaviours (“This coming month I want to eat healthy”). People may thus be “nudged” by temporal time marks.

In sum, temporal landmarks might impose disruption in consumers online purchase behaviour, such that consumers change their purchases pattern during this period. More specifically, temporal landmarks may activate healthy goals, and thus may lower the tendency to purchase vices purchases. Yet, these temporal landmarks and its effect on online consumption behaviour has not been tested in an online grocery setting.

16

(Peets & Wilson, 2013). Hence, consumers may change their online behaviour by purchasing more healthy food – and thus less vices – at the start of a new year:

H6a: The first month of the new year is negatively related to the SOV in consumers’ online grocery basket.

Moreover, Dai et al. (2014) found that people search more on the term “diet” on Google at the start of each calendar cycle: beginning of the week, month and year. Additionally, Ayers et al. (2014) suggested that people are most likely to think about their health on Mondays. Hence, starts of new calendar cycles may impose people to change their online purchase behaviour:

H6b: The first day of the week is negatively related to the SOV in consumers’ online grocery basket, compared to the rest of the week.

H6c: The days since the start of the month are positively related to the SOV in consumers’ online grocery basket.

Lastly, public holidays impose a temporal landmark (Dai, et al., 2014). These events can trigger the Fresh-Start-Effects and encourage people to pursue healthy goals. Therefore the following hypothesis is suggested:

H6d: The first week after a public holiday is negatively related to the SOV in consumers’ online grocery basket.

2.6 Vice Promotions influencing Vice purchasing

Given the importance of vice promotions in influencing consumers’ SOV, the impact of vice promotion cannot be ignored in the studying time effects on vice promotions. For this reason, vice promotions are additionally added to the theoretical framework.

17

has been tested in a physical store, but has not been applied to the online grocery channel. As promotion serves as an immediate economic incentive to purchase a product, vice promotions are presumably affecting consumers’ SOV online:

H7a: The number of vice promotions positively influences the SOV in consumers’ online grocery basket.

Baumeister (2002) proposes the daily cycle of life can be interesting for consumers’ self-control depletion. Failures in self-control are seldom seen in the morning, but become more likely during at the end of the day (Baumeister, 2002). Indeed, consumers that order at the end of the day before dinner, in a state of hunger are more likely to purchase vices (Loewenstein, 1996). Vice promotions are found to diminish consumers’ capacity for self-regulation by providing rewards (e.g. paying less, gaining credit) (Nakamura et al. 2015). Yan et al. (2017) notice that justification effects of vice promotions strengthen the preference for vice choices by depleting consumers’ self-control resources.

In sum, vice promotions deplete a person’s self-control, this effect is probably stronger at the end of the day as failures in self-control become more likely in the end of the day (Baumeister, 2002). Moreover, consumers that purchase groceries at the end of the day before having dinner are more likely to purchase vices (Loewenstein, 1996). Combining previous arguments lead to the following hypothesis:

H7b: Dinner time positively moderates the effect of the number of vice promotions on the SOV in consumers’ online grocery basket.

New Year's day is seen as a fresh start for new resolutions (Peets & Wilson, 2013). Hence, especially around this period people are (re)considering their behaviour patterns. As this public temporal landmark increase good intentions for the new year (i.e. eating healthy) (Dai, et al., 2014), vice promotions may have less effect on the amount of money spent on vices relative to the total amount of money spent on consumers’ shopping trip during this period. This rationale leads to the following hypothesis:

H7c: The first month of the new year (Fresh-Start-Moments) negatively moderates the effect of vice promotions on the SOV in consumers’ online grocery basket.

2.8 Control Variables

18

2.8.1 Week numbers. The focal online grocery company has special promotion weeks throughout the year

(to preserve business confidentiality, these promotional weeks are referred to as Promotion week 2 and

Promotion week 2). During these promotional weeks considerable incentives are provided to the

consumer (i.e. “Buy one product get one free”). These promotional weeks may affect the amount of money spent on vices relative to the total amount of money spent on consumers’ shopping trip. Hence, these promotional weeks are included in the model as control variables.

2.8.2 Day order delivered. The day of the week that an order is delivered may influence consumers’ SOV.

For example, if a consumer knows his/her groceries will be delivered on Saturday, (s)he may purchase more vices because friends will come Saturday. As the focus of this study is not on the delivery day, this effect will be controlled for.

2.8.3 Days since last delivery. Consumers that regularly order online may show different behavioural

patterns compared to consumers who ordered occasionally. Therefore, a control variable will be added which indicates the days since last delivery. Similarly to Milkman et al. (2010) a dummy variable will be included in the model indicating whether a consumer ordered more than 60 days ago.

2.8.4 Times opened from submitted. The number of times a consumer opens his/her order may influence

consumers’ SOV. For example, a consumer who opens his/her order several times to create his optimal order, would probably show less impulse behaviour as this will be corrected the next time (s)he opens the order. Hence, constantly changing an order may lead to an optimal rational order, which is less biased to time effects.

2.8.5 Days between webvisit order. If the time between making an order and delivery increases the

impact of time on SOV may be lower, because the consumer is making his/her decisions in several time moments and could therefore be less subjective to time factors.

2.8.6 Number of products and Monetary value promotional products. Similar to Milkman et al. (2010) the

numbers of products was included as a control variable. Also the monetary value of the promotional products in consumers’ basket may influence consumers SOV.

19

2.8.8 Seasonal Influences. During the summer/winter holidays consumers may purchase less groceries

online as they are on holiday. Additionally, consumers may purchase more unhealthy products during the winter months as several cultural events are celebrated around this time of the year. It is relevant to consider the fact that shopping patterns differ across seasons (Zhuang, Tsang, Zou, Li & Nicholls, 2006).

2.8.9 Promotional meal box. The focal online retailer offers a special meal box, which include a recipe and

the required groceries. These special meal boxes include more than one product. However, in the transactional data set this is counted as one product only. This may bias the results, and therefore it is relevant to keep this promotional meal box in the model.

2.8.10 Home Delivery and Pick Up Point. The focal online retailer offers two different channels through

which consumers can receive their online grocery orders: Home Delivery (HD) or Pick up Point (PUP). HD requires a minimal order vale of €70 and PUP has no minimal order value. Consequently, the size and price of these orders between the two different channels may differ which could influence consumers’ SOV.

2.9 Conceptual Model

The objective of this study is to identify the effect of different time effects on the amount of money spent on vices relative to the total amount of money spent during consumers’ shopping trip (SOV). Therefore, different hypotheses are proposed (Table 2), which are all visually summarized in the conceptual model (Figure 1).

20 Table 2 Summary of the hypotheses

Hypothesis Description Direction

H1 Online Lead Time has a negative effect on the SOV in

consumers’ online grocery basket.

-

H2 Online Grocery Shopping Maturity has a positive effect on the SOV in

consumers’ online grocery basket.

+

H3 Online Grocery Shopping Maturity has a higher impact on SOV in online

consumers’ grocery basket online compared to Online Lead Time.

+

H4 The end of the day (Late Evening) is positively related to the SOV in

consumers’ online grocery basket, compared to the rest of the day.

+

H5 Time before Meal (‘hunger state’) has a positive effect on the SOV in

consumers’ online grocery basket.

+

H6a The first month of the new year is negatively related to the SOV in consumers’

online grocery basket.

-

H6b The first day of the week is negatively related to the SOV in consumers’ online

grocery basket, compared to the rest of the week.

-

H6c The days since the start of the month are positively related to the SOV in

consumers’ online grocery basket.

+

H6d The first week after a public holiday is negatively related to the SOV in

consumers’ online grocery basket.

-

H7a The number of vice promotions positively influences the SOV in consumers’

online grocery basket.

+

H7b Dinner time positively moderates the effect of the number of vice promotions

on the SOV in consumers’ online grocery basket.

+

H7c The first month of the new year (Fresh-Start-Moments) negatively moderates

the effect of vice promotions on the SOV in consumers’ online grocery basket.

21

3. RESEARCH METHODOLOGY

This chapter describes the research methodology of this research. First, the data collection and the data purification process will be explained, followed by some descriptive information about the given data. In the section method of analysis, the specific used panel analysis will be described. The chapter ends with the model specifications and the operationalization of the variables of the model.

3.1 Data collection

The data presented in this research were collected from a large Dutch online grocery retailer. Its consumers can place orders by searching through the product collection the company’s website and consumers can then add products to an electronic shopping cart. When ordering, consumers have the option to choose from different delivery time slots, ranging from tomorrow to further in the future. There are two ways to receive a grocery order: by Home Delivery (HD requires a minimal order value of €70) and Pick Up Point (PUP has no minimal order value, €1.50 service cost). Prices of home delivery differ between €4.95 to €9.95, depending on the spread of the time frame (a large time frame is cheaper compared to a small time frame). For both these methods, payment must occur with a debit card. This equal payment method between the two different delivery options is important as previous studies illustrate that payment method has an influence on vices purchased (Thomas, Desai & Seenivasan, 2011).

The transactional dataset was aggregated to weekly transactional data. The transaction dataset included an unique and anonymous consumer id, a unique order number, the date of consumer first order, start moment date and time order, end moment date and time order, date of delivery, total order value, total number of products, total order value vice, total number of product vice, the monetary value promotional products, service method (PUP/HD), promotional meal box, order time in days, and days since last delivery. If a consumer opened the order several times before submitting, the data explained how many times the order was opened from submitting, as well as the first and last dates when the consumer modified his or her online order. Note that online shoppers could modify their selections after placing an initial order up until an cut-off time that certified the retailer to transport and deliver the groceries.

22

Banquet’. Together, these three categories form the online vice category. Nuts, Toast and Bakery were not considered to be part of the vice category. When the promotion file was completed, it was merged with the transactional data set.

This study was restricted to consumers who ordered their groceries for delivery between 1 and 29 days in advance, between January, 2015 and April, 2017. As one of the goals of this study is to find out what the effect of Online Maturity is on consumers’ SOV an initialization period of 2 years was used by following only those consumers who started ordering in 2015. The data contained information of all the consumers that started purchasing online groceries in 2015 (1.133.555 observations of 108.968 consumers). The B2B consumers were excluded from the analysis, resulting in 977.637 observations of 100.756 B2C consumers. Consumers who had an average SOV of zero were excluded from the analysis, as these consumers did not purchase SOV at all and thus show no dynamics in SOV. Indeed, explorative research shows that consumers with 100% vices were found to be outliers. These consumers may purchase vices for a special event. As we are interested in the dynamics in vice purchase over time from regular shopping trips, these observations are less informative. Similarly to Campo & Breugelmans (2015), consumers with 100% vice were removed from the dataset (Campo & Breugelmans focused on multi-channel shoppers, and these researchers removed those consumers with 0% or 100% multi-channel allocations). This allowed the researcher to study regular grocery trips of consumers over time. In the end, the dataset contained 876.887 observations of 65.886 consumers.

3.2 Data purification

Before performing the data analysis, the data was cleaned such that it was put in a format that was usable for analysis. The data showed some missing values for the sales variables (e.g. Monetary value

23

original variable. If a consumer opened the site only once for placing an order, the minimal value of number of Times opened from submitted should be at least 1. Hence, zeros should be ones, ones should be twos etc. Therefore, the variable was recoded to valid numbers by adding the number one to the variable. The variable number of virtue products contained 25 missing observations. In the original dataset these values were also missing. Following Kline (2011) less than 5% of missing values in a variable in a large sample is of little concern and listwise deletion is adequate. Since there 876.887 observations, deleting 25 data points (0,003%) would not be considered a problem. Also consumers’ first orders were removed as the focal online retailer found that first orders differ significantly from regular orders. Indeed, a t-test shows that consumers’ first order differs significantly: consumers order significantly less products (t(876.887)=73.5371, p=<.01) , spend less money on vice (t(876.887)=20.1813, p=<.01) and order relatively less vices (t(876.887)=16.0361, p=<.01). In the end, the data contained 857.793 observations of 59120 consumers.

Before estimating the model, the total sample was checked on outliers by using boxplots and histograms. The occurrence of outliers is applicable in panel data, as large panels of individuals are likely to include irregular observations (Bramati & Croux, 2007). Three consumers had an extreme order value (€2112.12), an extreme order promotional value (€1318.65), an extreme number of products (1246 products). Another outlier was found in the number of products of vice products: one person ordered 792 products of the vice category. Although the occurrence of outliers can strongly bias panel data estimators, by removing these outliers from the sample valuable information will be lost. Moreover, outliers can provide important information by increasing the variation in the explanatory variables (Wooldridge, 2013, p.316). Therefore, it was chosen to keep the outliers in the model.

3.3 Data description

24 Table 3 Summary Statistics and Variable Descriptives

Notation Name Description Mean SD Min Max

Spending (€) Total order value in € for consumer i in order t. 113.35 49.10 .25 2112.12

Spending

vice(#)

Total order value in € for vice products for consumer i in order t.

7.64 8.17 0 910.80 Share Of Vice

products (€)

Share Of Vice products in € of consumer i in order t.

.067 .0594 0 .9955 Number of

products

Number of products in order for consumer i in order t.

59.133 26.78 1 1246 Number of

vice products

Number of vices in order for consumer i in order t. 5.370 5.875 0 792 Share Of Vice

products (#)

Share Of Vice products in numbers of consumer i in order t.

.086 .073 0 .9897 Online Lead

Time

Days between last visit to confirm order and the date when the consumers’ groceries delivering by consumer i in order t.

1.739 1.499 0 29

Online

Maturity

Number of (previous) online shopping trips at grocery retailer by consumer i in order t.

20.646 18.19 2 178 Times opened from submitted

Continuous variable indicating how often consumer i opens the order after starting an order in order t. 2.193 1.842 1 142 Days between webvisits

Continuous variable indicating the difference in days between starting and ending an order of consumer i in order t.

1.241 3.080 0 169

Notes: time t denotes the sequential order moment of consumer i. For example: order t=4 means that a consumer places his/her fourth order since (s)he started ordering online at the focal online retailer.

3.4 Method of Analysis

The data used in this study is of numerical form, the research method therefore is quantitative. This study includes a quasi-experiment (i.e. natural experiment), because the subjects in this research have not been randomly pre-assigned to different treatment conditions as is happening in true experimental designs (William, Shadish, Cook & Campbell, 2002). Moreover, in this study the independent variables are not controlled by the author: the independent variables are naturally manipulated (i.e. a grocery shopper decides for him/herselves when (s)he wants to order groceries).

The advantage of a quasi-experiment is that it is very useful to investigate natural phenomena. This is interesting because research has shown that actual healthy behaviour differs from stated consumers’ health intention (i.e. the ‘intention-behaviour gap’) (Norman & Conner, 2005). Another advantage of quasi-experiments is its applicability to real world settings. One major limitation is that because of the deficiency in randomization, it is harder to rule out confounding variables. This introduces threats to internal validity (Harris et al., 2006).

An important type of quasi-experimental research design is a panel study (also known as longitudinal study) (Frensch, 2007). In a panel study, a selected group of consumers is repeatedly investigated across time series (Wooldridge, 2013, p.432). Panel data differs from pooled cross-sectional data in that the same cross sectional units are followed over a certain time period (Leeflang et al., 2015).

1 _{Estimation of the panel linear regression analysis models were done using panel linear model "plm" (Croissant and}

25

In this research the data was measured over multiple time periods for the same consumers, therefore panel data models were used. To analyse this, special techniques were required, e.g. it would be wrong to treat 100 consumers who shopped at 4 different points in time as though they were 400 independent observations. This can result in an error estimate that is too small. Hence, it cannot be assumed that the observations are independently distributed across time (Wooldridge, 2013, p.432). Panel data has numerous advantages over cross-sectional or time-series data. Panel data accounts for unobserved individual heterogeneity (Baltagi, 2013). Time-series and cross-sections methods which do not control for this heterogeneity have a risk of obtaining biased outcomes. In addition, panel data provides more variability, less collinearity among variables and more efficiency compared to cross-sectional or time-series data (Baltagi, 2013).

3.4.1 Panel models. Panel models can be divided in balanced and unbalanced panel sets. If the same t

time for each n individual is present, then the data is a balanced panel. Contrary, if the data sets have missing times for at least some individuals the data is unbalanced (Wooldridge, 2013, p.473). Economic panel datasets often happen to be unbalanced (Croissant and Millo, 2015). The used dataset contained an unbalanced panel dataset (Consumers=59.120, Observations=857.793, Consumers’ sequential order number=1-177). This panel thus does not have the same time periods for all individuals (Wooldridge, 2013,p.473). The applied panel regression method2 was able to deal with unbalanced panel datasets (Croissant & Millo, 2008).

Broadly speaking three different panel models can be defined: pooled regression, fixed effects and random effects (Greene, 2000). In the pooled regression model, one common constant term is included, which is pooled across individuals i and across time t. Ordinary least square (OLS) provides then provides consistent and efficient estimates of the pooled and the slope vector β. So, in the standard linear regression model ( = β ) the model assumes that the intercept and the slope coefficients are identical for all consumers and time periods. The error term varies over time and captures all

unobservable factors that affect . In this panel study, the same consumers are observed, and hence it

seems unrealistic to assume that the error terms for different time periods are uncorrelated (Wooldridge, 2013).

To statistically test whether a panel effect is present, the Breusch Pagan Lagrange multiplier (LM) was performed. This test helps to decide whether a random effect or a pooled estimation should be used. In the standard OLS regression model, the intercept is the same across individuals. The null hypothesis propose that the variances of the errors across units is zero, thus no significant differences between units and thus no panel effect:

2