Is bigger better? The influence of assortment variety on food waste.

(1)

Is bigger better? The influence of assortment variety

on food waste.

(2)

Is bigger better? The influence of assortment variety on food waste. By

HUGO SNIJDER

Completion date: 25th August 2015

Master Thesis

Msc Marketing, specialization Marketing Management and Marketing Intelligence University of Groningen, Faculty Economics and Business

First supervisor: Jenny van Doorn Second supervisor: Lara Lobschat

(3)

Abstract

This study investigates the influence of supermarket assortment size on the amount of avoidable food waste. Moreover, this study is the first to examine the moderating effects of individual’s health motivation and the type of shopping trip consumers usually undertake on the relationship between assortment size and the total amount of household food waste. By means of a food waste diary approach, the quantity and composition of food waste in 23 food categories is analyzed. The outcomes of this research show that consumers shopping at supermarkets with a larger assortment waste less. Main shopping trips and amotivation significantly increase the strength of the effects of assortment size on the total amount of food waste.

(4)

Management summary

Globally, a third of all food produced for human consumption is lost or wasted throughout the whole distribution chain (Gustavsson et al., 2013). The wastage of food has an enormous environmental impact, because huge amounts of resources used in the food production and distribution process are wasted. Households food waste is the largest single contributor to the huge amount of food waste. Besides the environmental impact, food waste also affects consumers financially. In the Netherlands, individuals yearly throw away 50 kilogram of edible food, which represents a value of 150 euro (Van Westerhoven, 2013). Therefore, the Dutch government has set the milestone to reduce food waste by 20 percent at the end of 2015 (Ministerie of Economische Zaken, 2013).

Nevertheless, very little research is conducted on the influencing factors on consumer’s waste behavior. Considering the potential benefits of reducing food waste, profound marketing research is necessary to gain deeper insights into avoidable food disposal behavior by consumers. Literature research indicates that buying too much food and being temped in stores are often named as main reasons of food waste. Since almost 80 percent of all food is bought in supermarkets and supermarket assortments can shape consumer preferences (Simonson, 1999), the influence of assortment size on food waste is examined into more detail. In addition, prior literature suggests the potential influence of the type of shopping trip consumers typically undertake and individual’s health motivations on the quantity of food waste. This study provides new insights regarding the influencing factors on total amount of food waste by focusing on marketing influences, consumer characteristics and shopping behavior. In this research, marketing influence is defined as the perceived assortment variety, while consumer characteristics are conceptualized as the different states of health motivation (e.g. autonomous motivation, controlled motivation or amotivation). Shopping behavior is defined as the type of shopping trip consumers undertake (e.g. main shopping trip of fill-in shopping trip).

In order to analyze the potential influence of these factors data is collected by a longitudinal food waste diary approach in combination with a background questionnaire. In total, 98 households participated in this study. The respondents were asked to record the quantity of the disposed products in 23 different categories. In addition, respondents recorded which supermarkets they visited and indicated whether the shopping trip could be labeled as a main shopping trip or a fill-in shopping trip. Since several variables are measured at a daily level, the structure of the data is referred to as panel data. Therefore, panel data analysis is most appropriate.

(5)

shopping trip type and health motivation, showed to have a significant influence on the relationship between assortment size and the quantity of food waste. First, the perceived variety at main trips showed as expected to increase food waste, while the effect of fill-in trips cannot be confirmed. Furthermore, the degree of a person’s amotivation showed to strengthen the effects of assortment size on the total amount of food waste. The two other states of health motivation, autonomous motivation and controlled motivation, did not show to influence the amount of food waste. This research is to the best of the author’s knowledge the first to investigate the combination of the previously described factors on the total amount of food waste. Therefore, this study helps to fill the gap in literature with regard to knowledge on influencing factors of food waste. Furthermore, this research has several implication. The trend in the Netherlands of supermarkets yearly increasing in size should satisfy public policy makers as long as the perceived assortment variety increases as well. Supermarket chains could contribute to the battle against food waste by making an effort to increase the variety perception of consumers in small sized supermarkets. Additionally, consumers making main shopping trips are more receptive the effects of an assortment. Therefore, it should be promoted to make main shopping trips at large supermarket, while fill-in trips to small supermarkets should be promoted. In the same vein, amotivation enhances the effect of assortment size on food waste. Consequently, amotivated consumers waste more food when they have shopped in supermarkets with small assortments. Therefore, policy makers should promote a healthy living style among amotivated individuals. Overall it can be concluded that in the battle against food waste ‘bigger is better’ in terms of supermarket assortments.

(6)

Preface

(7)

1. Introduction

One-third of the food produced for human consumption is lost or wasted globally, while hunger is still one the most urgent development challenges (Gustavsson et al., 2011). Inherently, this means that huge amounts of resources used in food production and distribution processes are wasted and lead to an unnecessary environmental impact (Williams et al., 2012). For instance, the global carbon footprint of wasted food is estimated to be twice the size of the total greenhouse gas emissions of all road transportation in the United States (European Union Committee, 2014). Besides ethical and environmental dimensions, food waste also affects consumers financially. Dutch consumers throw away 50 kilogram of edible food per capita each year, which represents a value of 150 euro (Van Westerhoven, 2013). Reducing food waste is relatively inexpensive, but the potential benefits are enormous (Parry, Bleazard, and Okawa, 2015). Food waste occurs throughout the whole distribution chain, from farm to fork, but household food waste is the largest single contributor to the total food waste (Graham-Rowe, Jessop, and Sparks, 2013; Koivupuro et al., 2012; Parfitt, Barthel, and Macnaughton, 2010; WRAP report, 2013). Therefore, the remainder of this article focuses on consumer food waste.

(9)

2 Focusing on what type of food is thrown away mostly, it becomes clear that a distinction between different product categories is necessary. A large differentiation exists in the wastage rates for different food types (Parfitt, Barthel, and Macnaughton, 2010). Perishable food items account for the highest proportion of food waste (Quested et al., 2013). For instance, in the UK are fresh vegetables and salads (19%), dairy and eggs (10%) and fresh fruit (8%) the largest contributors (in weight) to food waste (Quested et al., 2013). In the Netherlands are bread, vegetables, fruit and potatoes thrown away the most (Van Westerhoven, 2013). A distinction in food between virtues and vices could be drawn. Virtue and vice products are typically conceptualized in relation to each other as relative vices and relative virtues (Van Doorn and Verhoef, 2011). Vices provide an immediate pleasurable experience (e.g. the good taste of cake), but contribute to negative long-term outcomes (e.g. future health problems). In contrast, virtues are defined as being less gratifying and appealing in the short term (e.g. the less appealing idea of a wholesome sandwich), but have fewer negative long-term consequences (Wertenbroch, 1998). The products mostly thrown away (over 50%) are relative virtue groceries (Quested et al., 2013). On the other hand, relative vices are wasted way less (about 2%).

In order to contribute to a potential solution a deep understanding of the drivers of food waste is required. The subject of food waste has received some attention in academic literature and indicates several possible reasons for food waste, such as: a lack of plan or change of plan, buying too much, do not want to eat leftovers or do not know what to do with leftovers, failing to compile or comply with a shopping list, failing to carry out a food inventory before shopping, impulse purchases and high sensitivity to food hygiene (Lyndhurst, Cox, and Downing, 2007; Parfitt, Barthel, and Macnaughton, 2010; Stefan et al., 2012). In the UK, about half of the food waste is classified as food not used in time and up to 30 percent of food waste occurs because household served more food than could be consumed (Quested et al., 2012). The themes uncovered in these studies represent an important base of food waste, but a lack of understanding of the nature of food waste behavior still remains (Graham-Rowe, Jessop, and Sparks, 2014).

(10)

3 addition, retailers can influence household food waste through portioning, communication with customers but also through other marketing elements, such as price promotions or packaging (WRAP, 2015). Fortunately, the purchasing process is a well-documented subject in academic literature.

Since buying too much food and being temped in store are often named as one of the main reasons of food waste, the purchase process is of special interest (Lyndhurst, Cox, and Downing, 2007; Quested et al., 2013). However, buying habits in relationship with food waste is a less studied subject (HLPE, 2014). The purchase of food happens mostly in supermarkets (Cohen and Babey, 2012). In the Netherlands 77 percent of all food is bought in supermarkets (CBS, 2012). Therefore, the focus of this research is on the influence of factors driving behavior in supermarkets that ultimately lead to food waste. More specifically, the assortment of a supermarket is of interest because assortments can shape consumer preferences and influence whether and what consumers purchase (Simonson, 1999).

In this light, the distinction between virtues and vices is, besides the differences in the amounts of food waste, also valuable because consumers’ responses to products and assortments are influenced by the nature (e.g. virtue or vice) of products (Hui, Bradlow, and Fader, 2009). Research by Sela, Berger and Liu (2009) show that the size of an assortment influences what type of product people choose. These authors showed that larger assortments lead people to choose relative virtue products over relative vice products, because virtue products are easier to justify. If consumers have an accessible reason to indulge, however, larger assortments lead people to select vice products over virtue products. This might have serious implications for food waste, as virtue food is of higher risk of being wasted (Evans, 2011). Additionally, the sizes of supermarkets, and consequently assortment sizes, are yearly increasing in the Netherlands (CBS, 2012; DTZ Zadelhoff, 2011). What is the consequence of this trend for the chosen products, and more importantly, for the amount of food wasted?

(11)

4 might not stick to their intentions and do not consume (all of) their wholesome groceries, resulting in higher food waste (Graham-Rowe, Jessop, and Sparks, 2014). Another possibility is that less strongly health motivated consumer find reasons to indulge, which means that when shopping in a larger supermarket a higher share of vice products will be bought (Sela, Berger, and Liu, 2009). Hence, depending on the level of health motivation, different effects of assortment size on the amount of food waste are expected.

Second, the type of shopping trip is expected to influence the effect of assortment size on the quantity of food waste. Planning of shopping and shopping routines are important elements in avoiding food waste (Stefan et al., 2012). This is because people react differently to contextual cues depending on the type of trip (Bell, Corsten and Knox, 2011; Seetharaman, Ainslie, and Chintagunta, 1999). People usually make two types of visits to a supermarket. One for a large quantity of products when people stock up for a more lengthy period, and more frequent ‘fill-in’ trips when consumers purchase only a few items that they need immediately (Kahn and McAllister, 1997). Consumers are shown to be more receptive to in-store stimuli on major shopping trips (Kollat and Willett, 1967). For instance, they may react stronger to the effects of the size of an assortment to choose from. If major shopping trips lead people to be more receptive to in-store stimuli (e.g. such as assortment size) it might lead people to purchase a higher share of virtue products. Again, the higher share of virtue products bought does not automatically mean that all will be eaten (Graham-Rowe, Jessop and Sparks, 2014). Contrary, consumers on a fill-in trip are more purposeful and are therefore more closed-off to the shopping environment. As a consequence, the effect of assortment size on food waste might be weaker for these kinds of shopping trips.

Summarizing, this study seeks to investigate the effect of assortment size on food waste. Next to that, the influence of health motivations and type of shopping trip on the relation between assortment size and food waste is researched. This leads to the following problem statement: To what extent does assortment size affect the amount of food waste and how is this influenced by consumers’ health motivations and type of shopping trip?

In order to be able to answer this question, three research questions are identified: 1. How does assortment size influence the amount of food waste?

2. Do consumers waste more food when they shop in larger supermarkets versus smaller supermarkets depending on the type of shopping trip?

(12)

5 This study intends to contribute to literature in several ways. First, this study helps to fill the gap in the food waste literature from a marketing perspective. Several authors indicate that more research is needed to better understand the key factors that motivate, enable or prevent household waste (Graham-Rowe, Jessop, and Sparks 2015; Stefan et al., 2012). Most studies on household food waste concentrated primarily on identifying what food is most likely spoiled (WRAP reports 2009; 2010), who is most likely to waste food (Doron, 2012; Koivupuro et al., 2012; Lyndhurst, Cox, and Downing, 2007) and how people feel about food waste. However, little is known on the influencing factors on consumers’ disposal behavior. This study contributes to academic literature by investigating marketing influences1 and consumer’s shopping behavior2 on food waste. In addition, the possible influence of the rising desire to live healthy3 is taken into account. More specifically, to the best of the author’s knowledge, this study is the first to examine food waste caused by assortment size, individual’s shopping behavior and health motivations. As a result, this study extends the existing marketing and retailing literature.

Anecdotal evidence already hints at the influence of heath related activities (e.g. the wish of being a good provider by providing an abundance of healthy food, or the relationship between eating a healthy diet and reducing food waste) on food waste (Evans, 2011; 2012; Graham-Rowe, Jessop, and Sparks, 2014; Quested et al, 2013). However, none of these authors have studied the influence of health motivations driving behavior like this study does. Concluding, this study provides a more fine-grained perspective on the influencing factors on food waste.

The remainder of this article is organized as follow: the next section provides an overview of the previous literature. Then the conceptual model and the corresponding hypotheses will be introduced, followed by the methodology. Then the results will be presented, which lead to the discussion part. Finally, the conclusion of this study will be presented.

1_{In this study the marketing influence is operationalized as the size of a supermarket’s assortment.} 2

(13)

6

2. Literature review

This section discusses the theoretical underpinnings of food waste. A literature review of what already is known about food waste in general and it’s influencing factors, assortment size, health motivation and shopping goal, is discussed

2.1 Food waste

In developed countries, households are the main contributor to waste (Griffin, Sobal, and Lyson, 2009; Gustavsson et al., 2011; Parfitt, Barthel, and Macnaughton, 2010; Quested et al, 2013). Although food waste by households is not the only source of food waste, it is estimated to account for around 50% of the total amount of food wasted in the UK (Quested et al., 2013). The reasons why food is wasted can be linked to one or more steps in the food provisioning process, entailing food-related behaviors from purchasing , preparing, eating, to in the end disposing food (Stefan et al., 2012).

Overprovisioning of food is one of the main reasons of food waste (Evans, 2011; Lyndhurst Cox, and Downing, 2007). A study by WRAP (2009) in the UK revealed that about 40% of the food waste can be attributed to cooking, preparing or serving more food than can be consumed. Obviously, to be able to prepare more than can be consumed, in the first place too much food is bought by consumers. Another potential source of food waste is food not being used in time (WRAP, 2009), which is mainly driven by anxieties surrounding food safety and storage (Evans, 2011). From the report by WRAP (2009) could be abstracted that about 50% of the food waste is the result of food not used in time. The varying level of food knowledge is also named as possible factor driving an individuals’ propensity to waste food (Parfitt, Barthel, and Macnaughton, 2010; Graham-Rowe, Jessop, and Sparks, 2014).

Despite the enormous amounts of food waste, most people are waste aversive (Bolton and Alba, 2012). However, behaviors that generate waste are not necessarily waste-related (Quested et al., 2013). Stefan et al. (2012) found that the intention not to waste food did not transfer into less food waste. They argue that food waste may be perceived as mainly food-related behavior and thus being embedded in the daily food provisioning routines instead of conscious intentions. Another explanation may be that potentially conflicting personal goals are an obstacle in food waste reduction attempts (Graham-Rowe, Jessop, and Sparks, 2014).

(14)

7 prevent this action from happening has usually passed (Quested et al., 2013). For instance, buying food in abundance for dinner during a weekly shopping trip could lead to waste in a later stage, but by the time the food is spoiled it is obviously too late for preventive measures. Evans (2011; 2012) also argues for a more fine-grained perspective on how and why food gets wasted. Food waste often is the result of households performing everyday practices and deal with the contingencies of ordinary life (Evans, 2012). In other words, food wastage arises because of the ‘mismatch between the routines and rhythms of households and the food bought for consumption’ (Evans, 2011). Although Evans (2011) debates that household food waste cannot solely be attributed to individual’s thoughtlessness, the context of food purchasing (and consumption) is very important (Quested et al. 2013). Misunderstanding the context in which food is purchased and consumed will lead to sub-optimal efforts to affect behavior (Shove, 2010; Southerthon et al., 2011). Decisions made in the process of purchasing and preparing food influence the amount of food waste by consumers (Stefan et al., 2012). Being temped in stores is identified as one of the key explanations people give of food waste and it is inextricably linked to buying too much food (Lyndhurst, Cox, and Downing, 2007). The findings of these different authors highlight the importance to gain in-depth knowledge about the context in which consumers shop in relation to food waste.

Fortunately, a lot of research about in-store marketing is done (Neff, 2009; Bell, Corsten, and Knox, 2011; Cohen and Babey, 2012). In-store temptation can unfold in many different ways. It is highly important to understand the factors driving the extent to which consumers engage in in-store decision making (Inman et al., 2009). Store design and in-store factors are very important in determining sales (Cohen and Babey, 2012). Furthermore product assortments can play a key role in affecting buyer wants and preferences (Simonson, 1999).

2.2 Assortment size of supermarkets

(15)

8 product (Broniarcyzk and Hoyer, 2009). However, in the recent years, the general assumption ‘more is better’ has been questioned (Iyengar and Lepper, 2000; Van Ketel et al., 2003; Chernev, 2003; Broniarczyk et al., 1998). This stream of research demonstrates that assortment size is positively related to perceived assortment attractiveness, but that the effects are diminishing (Van Ketel et al., 2003), that too many options can lead consumers to not choose at all (Iyengar and Lepper, 2000) and that too many options result in more regret and less satisfaction with the options consumer choose (Schwartz, 2000). For instance, Iyengar and Lepper (2000) found that increasing the size of the jam assortment results in lower motivation to purchase jam.

The absence of a dominant option (Dhar, 1997; Redelmeier and Shafir, 1995) or clear prior preferences (Chernev, 2003) should lead to the occurrence of choice overload. Choice overload or overchoice could be best described as the negative impact of product assortment on consumer choice (Gourville and Soman, 2005; Scheibehenne, Grefeneder, and Todd, 2010). Although the size of the assortment is at the core of the choice overload hypothesis, there is no exact definition of what too much choice is (Scheibehenne, Grefeneder, and Todd, 2010). Assortment variety seems to be an important factor, as research showed that the variety offered influences which option consumers choose (Berger, Draganska, and Simonson, 2007). In addition, the quantity of consumption is influenced by the perceived variety of an assortment (Kahn and Wansink, 2004). Furthermore, perceived variety can determine consumers’ store choice (Van Herpen and Pieters, 2002).

Perceived assortment variety can be viewed as the function of assortment size and assortment structure (Chernev, 2011). First, larger assortments tend to be perceived as having more variety (Chernev, 2011). The perceived variety is moderated by number of distinct items compromising the assortment, the attractiveness of the assortment and the allocated total shelf space of the assortment (Broniarczyk, Hoyer, and McAllister, 1998). They found that retailers can make moderate reductions to the size of the assortment they carry, as long as this will not affect the perceived assortment variety. Smaller assortment can also reduce negative affect and regret and may increase consumer satisfaction (Dekimpe et al., 2011). Second, perceived assortment variety is a function of assortment structure, which refers to the organization and symmetry of items within an assortment (Kahn and Wansink, 2004). The rise of e-commerce, and more specifically online retail channels, has started the debate on the assortment size that (online) retailers should carry taking into account that consumer’s perceptions about assortment variety might have changed (Dekimpe et al., 2011; Elberse, 2008).

(16)

9 virtue products. They found that larger assortments tend to increase the choice difficulty consumers experience. Therefore, larger assortments shift consumers’ choice from vice products to virtue products, because these products are easier to justify. However, when consumers have an accessible reason to indulge the product choice shifts from virtue products to vice products for larger assortments. Situational factors can provide viable reasons to indulge.

This finding might have important implications for the influence of assortment size on food waste as it has been shown that the size of an assortment affects the type of product that will be chosen, which in turn might lead to more food waste. However, to the best of our knowledge this implication has never been subject of research until now.

2.3 Type of shopping trip

The variation in the type of shopping trip by a consumer is of growing relevance (Fox and Sethuraman, 2006) and especially how the type of shopping trip influences shopping behavior (Walters and Jamil, 2003). This is not surprising, given that about 70 percent of purchasing decisions are made in store (Vermeir and Van Kenhove, 2005). Moreover, research has shown that consumers are differently susceptible to in-store communication on different shopping trips (Nordfält, 2009).

(17)

10 Planning routines could also decrease product spoilage by preventing consumers to underestimate their inventory and buying products they already have at home (Chandon and Wansink, 2006). However, previous research has shown that the percentage of unplanned purchases is higher on fill-in shopping trips (Bell, Corstens, and Knox, 2011; Nordfält, 2009). Nordfält (2009) argues that this is the result of main shopping trips being more well-defined, while fill-in shopping trips are largely contingency-dependent constructions. This view questions the frequent assumption in marketing literature that the share of unplanned purchases increases with the size of purchases. Additionally, it has been argued by some authors that the type of grocery shopping trip moderates the choice of the type of store (Carpenter and Moore, 2006; Walters and Jamil, 2003). However, Nilsson et al. (2015) show that the type of shopping trip and choice of store format have a modest positive correlation, meaning that the traditional store format may not fulfil grocery shoppers’ needs. Many consumers do main shopping in small stores and fill-in shopping in larger supermarkets (Nilsson et al., 2015).

(18)

11

2.4 Health motivation

Another variable expected to influence the relationship between the size of a supermarket assortment and food waste is consumers’ health motivation. The desire to eat healthy is rising (Chandon and Wansink, 2007; Moorman and Matulich, 1993). Consumers report a growing interest in health and wellness (Food Marketing Institute, 2012). Moreover, concerns about healthy living and the best means to achieve it would seem to have reached a new, consistently high level (Petersen et al., 2010). Consequently, health motivated people strive to acquire food in line with their motivation (Moorman and Matulich, 1993).

(19)

(20)

13

3. Hypotheses

In this section is the conceptual model presented and the different hypotheses are formed.

3.1. Conceptual model

In order to investigate food waste and its drivers, further research is required. This study makes a contribution to scientific literature by examining the relationship between a supermarket’s assortment size and the amount of food waste. Furthermore, the moderating role of individual factors such as a consumer’s health motivation and shopping trip goal are investigated.

Figure 1: Conceptual model

The conceptual model used as basis for this research is presented in figure 1. As described in the literature review assortment size is expected to influence the amount of food waste and two factors are expected to affect this relationship, namely an individual’s health motivation and the type of shopping goal. The hypothesized relationships are more elaborated on in the subsequent sections. An overview of the expected relationships is presented in table 1.

Hypothesized signs variables Variables Expected

sign Rationale Source

Assortment size +

Selection of justifiable options when choosing from large assortment. Larger share of virtue

food leads to more food waste

Sela, Berger, and Liu, 2009

Shopping trip type

Fill-in -

Concrete shopping goal, closed-off from environment. Effect of assortment size on

food waste weaker.

Lee and Ariely, 2006; Morales et al., 2005; Huffmans and Houston, 1993; Walters and Jamil,

2003; etc.

Main +

Abstract shopping goal, more receptive to shopping environment. Effect of assortment

size on food waste stronger.

Health motivation

Autonomous -

Most likely to stick to intended behavior. Not only buy more virtue, but also consume

products resulting in less food waste. Gagné and Deci, 2005; Vansteenkiste, Lens, and

Deci, 2006; Ryan et al., 2008; Williams et al.,

1995; etc. Controlled +

Higher share of virtue products, but not able to stick to intended behavior resulting in

more food waste.

(21)

14 This study aims to better understand the development of food waste in households by investigating how assortment size influences consumer choices in a supermarket. Consumers take in the assortment they observe based on their shopping goal type (Huffmans and Houston, 1993). This is supported by Morales et al. (2005) who show that shopping goals affect the perceived variety of an assortment. Therefore, it is important to take consumers’ shopping goals into account when analyzing the impact of assortment size on food waste. In the same vein, the level of health motivation is expected to determine how consumers respond to the size of an assortment. Moreover, health motivation determines how likely it is one executes his/her intended behavior (Ryan, Deci, and Grolnick, 1995). Consequently, health motivation is an important factor in explaining consumers’ choices in supermarkets that ultimately lead to food waste. Furthermore, anecdotal evidence by Evans (2011) indicates that both health motivations and shopping behavior could play a role in existence of food waste. Based on these arguments are these variables selected in this study.

3.2. Assortment size and food waste

Buying too much food is one of the main reasons of food waste and is inextricably related to in-store temptation (Lyndhurst, Cox, and Downing, 2007). Assortments can shape consumer preferences and influence whether and what consumers purchase (Simonson, 1999). Specifically, assortment size influences option choice (Sela, Berger and Liu, 2009). Choosing from a large assortment is more difficult and as a result, consumer’s choice shifts to options that are easier to justify. Virtue products are easier to justify than vice products (Kivetz and Keinan, 2006; Kivetz and Simonson, 2002). This is mainly because the consumption of vice products typically is associated with guilt (Kivetz and Simonson, 2002; Khan and Dhar, 2006).

Consequently, consumers buy a higher share of virtuous products when they shop in supermarkets with a large assortment. This will in turn lead to more food waste, because virtue products are of a higher risk of being wasted (Evans 2011; 2012). This is because virtue products are usually perishable and need to be consumed within a certain timeframe. The short-shelf life and social-temporal demands of everyday life can lead to the disposal of food that was bought for consumption because household might be unable to consume the food within this timeframe (Evans, 2012).

(22)

15 perceived as relative vices are preferred over virtuous products, while the result is opposite when choosing from a large assortment. As a result, it is expected that the size of an assortment has a positive effect on the amount of food waste.

H1: Shopping in supermarkets with a larger assortment positively influences the amount of food waste.

3.3. Type of shopping trip and food waste

(23)

16 argues that the need to process external information is smaller when the shopping behavior is well defined and rehearsed. Therefore, the effect of assortment size on the amount of food waste is expected to be stronger for main shopping goal trips. These arguments lead to the following hypotheses:

H2a: A higher share of main shopping trips has a positive effect on the relation between assortment size and the amount of food waste.

H2b: A higher share of fill-in shopping trips has a negative effect on the relation between assortment size and the amount of food waste.

3.4. Health motivation and food waste

Health-related behavior, such as the desire to eat healthy, is often studied using self-determination theory (SDT; Ng et al., 2012). SDT is a general theory of human motivation (Deci and Ryan, 2000). According to SDT various types of motivation for behavior can be classified, namely amotivation, controlled motivation and autonomous motivation (Deci and Ryan, 2000). Central to this theory is the distinction of extrinsic motivation in autonomous motivation and controlled motivation (Gagné and Deci, 2005). This distinction is based on the degree to which extrinsic motivation is autonomous versus controlled. Autonomy refers to acting with the sense of volition and having the experience of choice (Gagné and Deci, 2005). Autonomous motivation thus involves the experience of volition and choice (Vansteenkiste, Lens and Deci, 2006). Controlled motivation refers to the sense of being pressured or coerced (Vansteenkiste, Lens and Deci, 2006). Both controlled and autonomous motivation is intentional (i.e. motivated), in contrast to amotivation, which involves a lack of intention and motivation (Gagné and Deci, 2005).

(24)

17 autonomously motivated) are most likely to stick to and perform intended behavior, such as buying and eating healthy food.

However, many people engage in behavior change because of controlled motivation (Ryan et al., 2008). Controlled motivation is usually based on an external reward. As lasting behavior change necessary for maintenance depends on accepting the regulation for change as one’s own, successful health behaviors and long-term maintenance would not result from controlled motivation (Koestner et al., 20008; Williams, 1996). Moreover, controlled motivation is largely unrelated to long-term adherence (Ryan et al., 2008). Reason to eat healthy based on controlling (e.g. because you would feel guilty if you didn’t) are thus expected not to result in the performance of the intended behavior. Amotivated people lack the intention to behave, and thus completely lack motivation and self-determination with respect to the target behavior (Ryan and Deci, 2000). Ryan and Deci (2000) argue that individuals are likely to be amotivated when they lack either a sense of efficacy or a sense of control with respect to a desired outcome. Accordingly, amotivation is associated with the poorest performance and mental-health outcomes (Ryan, Deci and Grolnick, 1995). Based on these reasons, amotivated people are likely to have no intention to perform healthy behavior. The underlying health motivation of consumers fit in the theory developed by Sela, Berger and Liu (2009). According to these authors choice difficulty increases when choosing from a larger assortment. However, the type of health motivation of a consumer is likely to determine the level of choice difficulty a consumer experiences. Furthermore, being able to exert self-control, that is, resist temptations, is a key factor the fulfillment of a goal (Metcalfe and Mischel, 1999). Autonomous motivation is accompanied by a feeling of self-control (Miquelon, Knäuper and Vallerand, 2012). Thus, autonomous health motivated consumers are likely to select virtuous food in most circumstances and are unlikely to be tempted to buy relative unhealthy vice products. When choosing from a larger assortment, consumers have even more options in more categories to select virtuous food, which will lead to an even higher share of virtue food (Fasolo et al., 2009). Thus, these customers are not expected to experience a lot of choice difficulty (Kahn et al., 2014). Furthermore, autonomous health motivated consumers are not only likely to buy a higher share of virtue food, but also perform behavior in line with the desired outcome and thus consume this food (Ryan et al., 2008). Consequently, autonomous health motivation is expected to mitigate the relationship between assortment size and food waste.

(25)

18 more likely to experience difficulties in resisting counterproductive behavior, such as preferring to eat a delicious but unhealthy meal while the ingredients for a virtuous meal are already at home (Miquelon, Knäuper and Vallerand, 2012). Thus, behavior based on controlled health motivation leads to the purchase of more virtue food when shopping in large supermarkets, but at the same time, the intention to live healthy is not always translated into consuming all virtuous food. Indeed, past research has shown that the correlation between intentions and behavior is often low (Gollwitzer and Sheeran, 2006). As a result, controlled health motivation is expected to strengthen the relationship between assortment size and food waste.

Amotivated consumers do not have the intention to eat healthy. When choosing from a larger assortment a reason to justify vice food is more likely to be readily available, resulting in a higher share of vice products. For amotivated people it is easier to rationalize these vice purchases, which allow them to enjoy the immediate pleasure of consumption (Mishra and Mishra, 2011). The relationship of assortment size and food waste is likely to be less strong for amotivated consumers. These arguments result in the following hypothesis:

(26)

19

4. Methodology

In order to answer the research question data is collected in two ways, namely by means of a food waste diary (reported in appendix B) and a questionnaire (see appendix A). A food waste diary approach is in line with methods used by Koivupuro et al. (2012) and Silvennoinen et al. (2014). Data collection took place in April and the beginning of May 2015. Both the food waste diary and the questionnaire were pre-tested before sending out the study, which led to several improvements in the research design and increased the clarity. To improve response rates 5 dining vouchers à 20 euro were randomly assigned to respondents who filled in the study completely and insights into respondents’ waste behavior were offered afterwards. Respondents were also allowed to anonymously hand in the study.

The main reason to use a food waste diary approach was to collect information about respondents’ waste behavior. Respondents recorded the quantity of food waste in 23 categories on a daily basis for a period of three weeks. Participants had to weight or estimate the amount of food they wasted. To facilitate the estimation of the amount of waste a weight example of a typical product for each category was given. Food that could not be categorized in one of the 23 categories could be described in an additional category ‘others’. The categories used generate about 80% of all purchases of food (Van Doorn and Verhoef, 2014). Food categories could be further categorized in virtue (e.g. fruit, vegetable or bread), vice (e.g. alcoholic drinks, cookies or sweets) or neither (e.g. rice, pasta or meat) food, based on the distinction by Hui, Bradlow, and Fader (2009). An overview of the categories with products included can be found in table 2.

Food categories

Virtue category Bread, Cereals, Dairy products, Eggs, Fruit, Juice, Soup, Vegetables

Vice category Alcoholic drinks, Chees, Chocolate, Cookies, Crisps, Desserts, Nuts, Soft Drinks, Sweets

Neither category Chicken, Fish, Meat; Coffee, Tee; Dressings, Toppings; Rice, Pasta, Potatoes

Table 2: Subdivision of food categories

Besides recording food waste, respondents were asked to record in the diary when they visited which supermarket. Furthermore, respondents filled-in what kind of shopping trip they made, e.g. main shopping or fill-in shopping. For the ease of use, a legend was provided in which respondents could link a supermarket to a number (i.e. 1 = supermarket X in place Y, 2 = supermarket Z in place Y, etcetera). Hereafter, this number could be used in the diary to record when which type of shopping trip was undertaken.

(27)

20 the accuracy and truthfulness of food waste diaries. Second, by recording food waste in a diary awareness of waste might be increased, which already might have an influence on the amount of food waste.

A questionnaire is used to collect additional information, such as households’ socio-demographics and economic situation, the motivation to eat healthy and households’ perceptions about the variety of the assortment of supermarkets they visited during the observation period. The complete questionnaire is attached in appendix A.

To enhance insights in the data the daily observations are aggregated to a weekly level. For instance, food waste in the different categories is collected on a daily basis but transformed to a weekly level. This is because households did not waste food on a daily basis in many categories resulting in lack of data on a daily level.

The perceived assortment variety is measured using 7 different questions with regard to the assortment of the supermarket households visited. A scale developed by Broniarczyk, Hoyer, and McAlister (1998) is used to measure the perceived variety of a supermarkets’ assortment on a 7-point Likert scale. The 98 respondents have visited and evaluated 258 supermarkets in total. A correlation analysis shows that the different question with regard to variety all significantly correlate (p <0.001). However, question 4 (this supermarket offers a lot of variety of price ranges to choose from) correlates not as strong (0.265 < r < 0.447) with the other questions as the other questions do (r > 0.563). A reliability analysis is conducted to examine internal consistency using Cronbach’s alpha. The Cronbach’s alpha should exceed a value of 0.7 to allow summing the questions together (Nunnally, 1978). The scale for perceived variety is reliable, since α=0.877 for the 7 items. However, if question 4 is deleted, the Cronbach’s alpha increases to a value of 0.922. Based on the reliability analysis and the correlation analysis all questions except question 4 are summed into one variable per supermarket.

(28)

21 follows: the sum of the perceived variety on shopping trips in a week divided by to total number of shopping trips.

The size of a supermarket is a proxy for assortment variety (Levy and Weitz, 2004; Sinigaglia, 1997). Therefore, data on supermarkets’ floor area is collected to validate consumers’ variety perception. Distrifood Dynamics provided data on supermarkets’ floor area. In total, respondents used 119 unique supermarkets with an average floor area of 1459 m2 (SD=1304.72). The smallest supermarket is only 85 m2, while the largest supermarket is 12500 m2. Information about 6 supermarkets was unavailable, namely 2 local supermarkets, two supermarkets of the Aldi chain and two supermarkets of the Lidl chain. The missing data of the supermarkets of Aldi and Lidl are substituted by the sample mean of the chain and for the other two is the sample mean of all supermarkets used. The mean of Aldi in this study is 821 m2, for Lidl this is 1118 m2, and the mean of all supermarkets is 1459 m2. In the Netherlands, a supermarket has a floor area of 1700 m2 on average (DTZ Zadelhoff, 2011). The average floor area in the sample is slightly smaller than the national average, which might be caused by the underrepresentation of supersize supermarkets, such as the Albert Heijn XL format. Table 3 provides more details on the floor area and market share per chain used in this study and the national average.

Chain Sample average

m2 Sample market share

National average m24 National market share5 Lidl 1118 13,7% 1116 9,7% Aldi 821 7,7% 865 7,4% Albert Heijn 1399 34,2% 1633 34,1% Jumbo 1735 19,7% 1471 19,8% Superunie6 1113 17,9% 975 29%

Other7 2569 6,8% n.a. n.a.

Table 3: Overview of floor area per chain

In order to determine whether a supermarket’s floor area and the perceived assortment variety both measure the size of an assortment a Pearson correlation test is performed. The floor area of a supermarket correlates positively with how consumers perceived the variety of an assortment (p <0.01; r=.246). However, the effect is relatively small, indicating that floor area and the perceived variety measure not exactly the same construct. Important to note is that in this study small supermarkets not necessarily are convenience stores (such as AH to GO), but smaller sized supermarkets. In the same line, large supermarkets are not necessarily supersize supermarkets (such

4

The average floor area per chain for the year 2014 is obtained from Van Lit (2014). 5_{The national market share for the year 2014 is obtained from Nielsen (2015).} 6

(29)

22 as Jumbo XL or Albert Heijn XL), but relatively large sized supermarkets.

As described in the hypothesis section different types of health motivations are measured using the Treatment Self-Regulation Questionnaire (Ryan and Connell, 1989). The TSRQ was developed to assess various health behaviors (Levesque et al., 2007). Questions related to autonomous motivation, controlled motivation and amotivation are measured on a 7-point Likert scale. The subscale for autonomous motivation consists of 6 items and is internal consistent, because the Cronbach’s α= 0.73 and thus exceeds the critical value of 0.7. Next to that, the correlations between each item and the total score from the subscale are larger than 0.3, which is the critical value (Field, 2010). The subscale for controlled motivation consist of 6 items as well and exceeds the critical value of 0.7 only by deleting question 7 ‘The reason I eat healthy is because I would feel bad about myself if I did not eat a healthy diet’. Removing this item improves Cronbach’s α from 0.639 to 0.703. Lastly, the subscale amotivation shows moderate internal consistency, since Cronbach’s α is only 0.621. The scale could not be improved by deleting an item. Although the Cronbach’s α does not exceed the critical value, the subscale for amotivation is maintained because it has proven its value in a validation study by Levesque et al. (2007) where the internal consistency proved acceptable (α>0.73).

Finally, to verify whether the questions used represent the structure of health motivation as described by Ryan and Connell (1989) a principal component analysis (PCA) is performed. In a principal component analysis the linear components (factors) of the correlation matrix are calculated by determining the eigenvalues of the matrix. These eigenvalues are used to calculate eigenvectors, the elements of which provide the loading of a particular variable on a particular factor (Field, 2010). A fixed number of extracted factors are used to represent the motivational statuses. Varimax orthogonal rotation is used to improve interpretation. Most items (9 out of 14) load on the correct subscale (see appendix C). Concluding, the subscale of TSRQ demonstrates acceptable reliability.

(30)

23

Variables Subdivision Cronbach’s α Mean Std. Dev.

Perceived variety Small 0,922 3,660 1,014

Large 6,114 0,534

Shopping trip type8 Fill-in n.a. 0,468 0,329

Main n.a. 0,532 0,329

Health motivation

Autonomous 0,73 5,740 0,872

Controlled 0,703 2,600 1,101

Amotivation 0,621 2,043 1,194

Table 4: Overview variables

4.1. Descriptive of the collected data

In total, 97 household from the Netherlands and 1 from Germany participated in this study. The 98 household account for 221 persons of which 57 are children. However, from the 98 participants 3 did not fill in the food diary correctly. These respondents are removed from the database. The average household in this study consists of 2.26 persons of which 0.58 are child. From the participating household, 25 are single households, 36 are two-person households and 34 are households of more than two persons. Furthermore, 61 households have no children. The youngest respondent is 20 and the oldest 82, but respondents are on average 46.29 years old. From the respondents 83.7 percent is female and 16.3 percent is male. The largest share (36.7%) of respondents stated ‘higher professional education9’ as highest completed education, followed by intermediate vocational education10 (25.5%), university (18.4%), MAVO or VMBO (11.2%), HAVO (6.1%) and VWO (1%). Nearly 67 percent of the respondents are employed, about 10 percent is student, 4 percent is houseman or housewife, another 4 percent is unemployed, 4 percent reported ‘other’ as occupation and 11.2 percent of the respondents is retired. Employed respondents work on average 30 hours a week, with a minimum of 4 hours a week and a maximum of 60 hours a week. The main category of a monthly household’s income of all members in this study is between 3000 euro and 3999 euro with 21.4%, followed by between 2000 and 2999 euro with 16.3 percent and 4000 euro or more with 14.3 %. In total, 13 participants preferred not to say what their monthly net income was.

The frequency of shopping in a week could be inferred from the food waste diary, but is also verified in the questionnaire. The largest share of respondents (32.7%) indicate to shop 3 times per week on average, while 24.5% shops 2 times per week and 17.3% shops 4 times per week or more. Based on the food diary, respondents reported an average shopping frequency of 3.13 times per week with a standard deviation of 1.70, which is in line with the results from the questionnaire. In total, 872 visits to supermarkets have been made in the observation period, from which 43.5% was at a supermarket with a small assortment and 56.5% at a supermarket with a large assortment. Main

8

Measured as a ratio: fill-in or main as a fraction of total shopping trips 9_{Higher professional education is HBO in Dutch}

10

(31)

24 shopping trips account for 49.4% of all shopping trips, while fill-in trips account for 51.6% of all shopping trips. Almost 40% of all fill-in shopping trips were made a small assortment supermarket. More than half of the main shopping trips were made at supermarkets with a large assortment. Full details are provided in table 5.

Variables Minimum Maximum Sum Mean Std. Deviation

Total visits to supermarkets 0 10 875 3,13 1,703

Frequency small assortment 0 7 381 1,37 1,511

Frequency large assortment 0 9 494 1,77 1,727

Main shopping trip 0 9 432 1,57 1,339

Fill-in shopping trip 0 7 443 1,61 1,460

Small assortment and fill-in trip 0 5 174 0,62 1,003

Large assortment and main trip 0 7 231 0,83 1,147

Table 5: Descriptives shopping behavior

Research by Lyndhurst, Cox, and Downing (2007) showed that one of the main reasons of food waste is that consumers buy too much food. To gain additional insights in the drivers of overprovisioning respondents indicated what their main reasons were in case of buying too much (see appendix E). The most named reasons for buying too much are large package sizes and being afraid of not buying enough with each 23.5 %. Additionally, respondents were directly asked what the main reason for food waste was. The most frequently indicated reason was ‘it was over the expiration date’ with 38.57% followed by ‘prepared too much’ with 36.43%. Surprisingly, the reason Lyndhurst et al. (2007) indicated as one of the most important drivers of food waste, namely ‘I bought too much food’, accounts only for 7.86 percent.

Focusing on food waste, table 6 illustrates the amount of food wasted per week, on average, and total for different categories. Because drinks are measured in milliliters and food in grams, the total amount of food waste is calculated in units. In total, 261.934 units of avoidable food are wasted with a mean of 87.311 units in total. Households in the sample wasted on average 909 units per week. Most food wasted is virtue food with 49.9 %. Vice food only accounts for 9.8% of the food waste, while neither accounts for 40.5% of the total food waste per week on average. Only 4 respondents did not waste anything at all.

(32)

25 kg11). Although this study tried to minimalize social desirable responses by providing the option to anonymously return the food diary, this bias still might have played a role resulting in lower recorded amounts of food waste. Another possible explanation is that households did not record everything they threw away or underestimated the amount food thrown away.

Type of

waste Week Total Minimum Maximum Mean

Standard Deviation Share of waste Virtue food waste 1 52134 0 3755 543 674 53,4% 2 44809 0 4810 467 645 52,3% 3 34484 0 1480 359 392 43,9% Week average 43809 456 49,9% Total 131427 Vice food waste 1 7971 0 1500 83 203 8,2% 2 9495 0 1125 99 206 11,1% 3 7975 0 1330 83 201 10,2% Week average 8480 88 9,8% Total 25441 Neither food waste 1 37540 0 2420 391 573 38,4% 2 31427 0 1960 327 440 36,7% 3 36369 0 2450 379 548 46,3% Week average 35112 366 40,5% Total 105336 Total food waste 1 97644 0 6825 1017 1113 2 85731 0 5010 893 971 3 78558 0 4155 818 845 average 87311 909 Total 261934

Table 6: Food waste per category

The upper part of table 7 provides an overview of waste in drinks and the lower part provides an overview of waste of food. A few food and drink categories contribute relatively much to the total. For instance, the categories coffee and tea and dairy products account for 388.91 ml and 342.17 ml, while fruit (282 gram), vegetables (318 gram), bread (253 gram) and leftovers (289 gram) is wasted the most of the food categories. The national average share per category is reported as well. Unfortunately, the study by Van Westerhoven (2013) used not the same subdivision of food categories. Consequently, for most categories is no information available. Comparing the categories for which information is available, some differences are noted. The share of dairy products in this study is twice as high as the national average, while the opposite is true for bread, fish and meat. The share of rice, pasta and potatoes is even three times higher in the study of Van Westerhoven.

11

(33)

26

Category Minimum Maximum Total Share Average

amount12 Mean per household Std. Deviation Alcohol 0 1750 8865 3,34% n.a. 90,46 282,232

Coffee and tea 0 5770 38113 14,38% n.a. 388,91 974,485

Dairy products13 0 5000 33533 12,65% 269,69 342,17 723,740

Desserts 0 400 1780 0,67% n.a. 18,16 59,023

Dressings and toppings 0 930 6030 2,27% 333,90 61,53 147,766

Soup 0 750 4136 1,56% n.a. 42,20 115,226

Fruit juices 0 500 4740 1,79% n.a. 48,37 113,609

Soft drinks 0 1500 4235 1,60% n.a. 43,21 174,353

Bread 0 1691 24793 9,35% 809,07 252,98 368,397

Cereals 0 150 320 0,12% n.a. 3,27 17,396

Cheese 0 575 3518 1,33% 77,05 35,90 80,233

Fish and meat 0 1750 9744 3,68% 404,53 99,43 207,790

Chocolate 0 200 430 0,16% n.a. 4,39 28,495

Cake 0 860 5471 2,06% 166,95 55,83 143,656

Chips, nuts and sweets 0 600 2524 0,95% 64,21 25,76 87,355

Eggs 0 500 4069 1,54% 38,53 41,52 86,091

Fruit 0 1906 27640 10,43% 552,22 282,04 382,973

Rice, pasta and patatoes 0 2100 20209 7,62% 911,80 206,21 303,133

Ready made meals 0 500 3020 1,14% n.a. 30,82 87,995

Leftovers 0 4800 28280 10,67% 77,05 288,57 605,325

Vegetables 0 2300 31128 11,74% 590,75 317,63 428,174

Baby food 0 190 190 0,07% n.a. 1,94 19,193

Others 0 450 2310 0,87% 154,11 23,57 72,186

Table 7: Overview of waste per food category

In addition, correlation matrices for all variables are produced using Pearson correlation coefficients. An overview of all correlations is provided in appendix D. Virtue waste correlates positively with vice waste (p <0.01; r=.272), neither waste (p<0.01; r=.283), and unsurprisingly with total waste (p<0.01; r=.817). Vice waste correlates positively neither waste (p<0.01; r=.267) and total waste (p<0.01; r=.518). Furthermore, neither waste correlates positively with total waste (p<0.01; r=.749).

Concentrating on shopping behavior and waste, a negative relation exist between supermarkets with large assortments and neither waste (p<0.01; r=-.174) and total waste (p<0.01; r=-.160). This indicates a lower share of neither waste and total waste for households going more often to supermarkets with a large assortment. On the contrary, a positive relation exists between small assortment and neither waste (p<0.01; r=.184) and total waste (p<0.05; r=.151). Furthermore,

12

The average amount in the Netherlands, based on research by Van Westerhoven (2013).

(34)

27 main shopping trips correlates negatively with virtue waste (p<0.05; r=-.139). There are no significant relations between the type of shopping trip and the perceived assortment variety. Focusing on health motivations, vice waste correlates negatively with the level of autonomous motivation (p<0.05; r=-.145), meaning that a person with a higher degree of autonomous motivation produces less vice waste. The same accounts for neither waste (p<0.01; r=-.206) and total waste (p<0.05; r=-.149) for autonomous motivated people. Controlled motivation has a negative relation with neither waste as well (p<0.01, r-.176). On the other hand, amotivation has a positive relation with vice waste (p<0.01; r=.194), suggesting that amotivated people produce a higher share of vice waste. Furthermore, controlled motivation correlates positively with amotivation (p<0.01; r=.504) and has no significant relation with autonomous motivation. Autonomous motivation correlates negatively with amotivation (p<0.01; r=-.359).

Zooming in on the demographics and food waste, age has a significant negative correlation with virtue waste (r=-.130) and a positive correlation with total waste (r=.212). The older the people are, the more they waste in total, but this is not because of virtue food waste. Gender and education have no significant correlation with food waste. Working hours per week correlate positively with virtue waste (r=.209) and total waste (r=.117). Household size correlates positively as well with virtue waste (r=.282), neither waste (r=.236) and total waste (r=.306). In addition, the number of children in a household correlate positively with virtue waste (r=.327), neither waste (r=.341) and total waste (r=.388). Lastly, income is positively related to all categories of food waste.

Some of the demographic variables are recoded in order to keep to model parsimonious. The variable occupation is transformed from categorical into a dummy variable, e.g. working versus non-working. Although income and education are strictly ordinal variables they are considered as continuous variables in the model. Results with regard to these variables are treated with caution because the intervals of these variables are not completely equal.

4.2. Method

(35)

28 A more flexible approach compared to AN(C)OVA and regression models is the panel data approach (Baltagi, 2008). The usage of panel data model has several benefits, such as the ability to control for individual heterogeneity. Studies not controlling for this heterogeneity could obtain biased results (Baltagi, 2008). Variables can vary with states and time or can be invariant. An example of a state-invariant variable is nationwide advertising on television or federal regulations, which does not vary across states. Variables could also be time-invariant, such as education or religion. Both education and religion are unlikely to vary much over time for an individual. Panel data are able to control for these state- or time-invariant variables. Furthermore, panel data deals effectively with multicollinearity issues, whereas regular time-series models are typically disturbed by multicollinearity (Baltagi, 2008). Next to that, panel data allow to decompose variation into variation between states of different sizes and characteristics and variation within states. Therefore, panel data methods give more informative data and more variability and more reliable parameter estimates are obtained (Baltagi, 2008).

Baltagi (2005) describes that the equation for a panel data regression differs from a regular regression in the double subscript on the variables within the panel data regression:

(1.1)

with denoting households and thus the cross-sectional dimension and describes time and thus the time-series dimension. Both subscripts are included for the dependent variable ( ), the explanatory variables ( ) and the disturbance term ( ). Most panel data studies apply a one-way error component model for disturbances:

(1.2)

where stands for the unobservable individual specific effect not accounted for in the model and is time-invariant, and accounts for the remainder of the disturbance and varies over both

individuals and time. In general two different techniques to analyze panel data are classified, namely fixed effects (FE) models and random effects (RE) models. The classification is based on the assumption of the error term.

FE models assumes the estimation of fixed parameters for and the remaining error to

(36)

29 explanatory variables on the de dependent variable. Therefore, FE models are not able to estimate the effects of time-invariant variables and are as a consequence omitted in the analysis. RE models assumes to have a random distribution and to be independent of the .

Additionally, the explanatory variables are assumed to be independent of the error term, both and

, for all households over time. This type of models is an appropriate specification for household

panel data where respondents are randomly drawn out of a large population (Baltagi, 2008), such as in this study. Unlike the FE model, a RE model is able to estimate time invariant variables because of the assumption that the entity’s error term is uncorrelated with the explanatory variables. A Hausman test is used to analyze if a FE model or a RE model is more suitable. The rationale behind this test is that for FE models a household’s error term and the constant (e.g. captures individual characteristics) should not be correlated with the others (Torres-Reyna, 2007). The null hypothesis indicates no correlation between the individual effects and . Significant results of the

Hausman test suggest that a FE model has to be applied, otherwise a RE model should be preferred. However, the Hausman test only differentiates between FE models and RE models. If a FE model is the preferred model, another test has to be conducted to justify the selection of a FE model over a simple pooled OLS. In order to apply the FE model a test of joint significance should be performed using the F-test (Baltagi, 2008). The null hypothesis is that all of the fixed intercepts are zero. Rejection of the null hypothesis indicates that fixed effects method should be applied. Similarly, an additional test should indicate that a RE model is more suitable than a pooled OLS. The Lagrange multiplier (LM) test is used to analyze whether a RE model or a simple OLS regression fits best to the data (Baltagi, 2008). The null hypothesis in this test is that variances across entities are zero. Rejection of the null hypothesis suggests that RE models is most appropriate. If the null hypothesis could not be rejected there are no significant differences across individuals, which suggest that a simple OLS regression could be performed.

First, to decide between a FE model and a RE model a Hausman test is performed. The Hausman test shows insignificant results (p=0.902), which indicates that a RE model should be preferred over a FE model. Second, the LM is conducted to indicate whether a RE model is more appropriate compared to pooled OLS. The Breusch and Pagan LM test shows that significant differences across households exist (p<0.001). Therefore, a RE model is the most appropriate model for the available data. The RE model is estimated using Generalized Least Squares. In addition, models are also estimated using maximum likelihood estimation (MLE) in order to be able to compare models based on the log likelihood. The MLE method has as an additional assumption for RE models that follows a normal distribution. Estimating RE models using MLE is mathematically

(37)

30 effects are hypothesized for shopping trip type and the different health motivations, the main effects are included as well. Interaction terms are normally included in a model by adding the product of the explanatory variables hypothesized to moderate the relation as an additional variable (Leeflang et al., 2015). The complete model becomes:

where stands for total food waste for a household in week . Assortment variety describes the

perceived variety of a supermarket’s assortment measured as the average perceived assortment of a supermarket for household in week . Furthermore, Share of main shopping trips describes the fraction of main shopping trips of the total shopping trips in a week and is measured as a ratio of the number of main shopping trips to total shopping trips. Controlled motivation, Autonomous motivation and Amotivation represent the degree of health motivation and vary for households, but are time-invariant. In addition, several interaction effects are included in the model, such as Assortment variety at main shopping trips and Assortment variety at fill-in shopping trips which describe the average perceived assortment variety for the type of shopping trip for household in week . The health motivation variables are multiplied with the perceived assortment variety to account for possible interaction effects for household in week . Likewise, the demographic and socio-economic variables Working hours (number of working hours in a week), Household size (number of persons in a household), Number of children (number of children in a household), Income (household income for all members), Age (age in years), Gender (0= female; 1= male), Education (highest finished degree) and Occupation (0= unemployed; 1= employed) are time-invariant and therefore only vary over households. In addition, measures variability in intercepts and is the

Is bigger better? The influence of assortment variety on food waste.