Softening consumers’ ecological footprint: preferences for meat and meat substitutes

(1)

Softening consumers’ ecological footprint:

preferences for meat and meat substitutes

Klasina Holthuis

(2)

SOFTENING CONSUMERS’

ECOLOGICAL FOOTPRINT:

PREFERENCES FOR MEAT AND

MEAT SUBSTITUTES

Klasina Holthuis

Faculty of Economics and Business Department of Marketing

Master thesis for MSc Marketing Management & Marketing Intelligence Completion date: January 11th, 2016

Address: Damsport 437, 9728 PS Groningen Telephone: +31645116835

Email: k.holthuis@student.rug.nl Student number: s2198762

First supervisor: Dr. F. Eggers Second supervisor: Dr. J. van Doorn

(3)

Abstract

With the current food consumption patterns, which are largely focused on meat and dairy, we will never be able to feed the 9 billion people in 2050 sustainably. The vast majority of forecasts indicate that lowering meat consumption is essential if we want to meet future food supply in a sustainable way. In this research, the preferences for meat and meat substitutes are investigated among a sample of 92 Dutch consumers. Choice-based-conjoint (CBC) analysis is used to discover what attributes consumers find important when ordering a hamburger at a restaurant. In order to reduce the different forms of bias associated with hypothetical

(4)

Preface

I would like to dedicate my master thesis to my dad, Willem Holthuis. Although not

(5)

Overview of figures and tables

Figure 1: Conceptual model ... 12

Table 1: Attributes and levels ... 14

Figure 2: Exemplary choice set ... 16

Figure 3: Descriptives ... 18

Table 2: Health concern factors ... 20

Table 3: Environmental concern factors ... 20

Table 4: Goodness of fit comparison ... 22

Figure 4: Attribute importance ... 23

Table 5: Average choice shares compared to no-choice option ... 23

Figure 5: Model specifications and corresponding utility estimates ... 24

Table 6: Utility estimates main effects ... 25

Figure 6: Model specifications and corresponding utility estimates ... 26

Table 7: Goodness of fit comparison ... 27

Table 8: Utility estimates main effects and interactions ... 28

Table 9: Summary of results ... 29

Table 10: Selection criteria ... 30

Table 11: Utility estimates per segment ... 30

Table 12: Summary of factor analysis results ... 50

(8)

2

1 Introduction

The challenge is real: in 2050, 9 billion people need to be fed (Godfray et al., 2010; Steinfeld et al., 2006). With the current food consumption patterns, which are largely focused on meat and dairy, especially in the Western society, we will never be able to accomplish this in a sustainable way. Increasingly, livestock production is being linked to many diverse problems in today’s world. Concerns are being raised about environmental sustainability, health and safety, and animal rights and welfare (Pluhar, 2010; Ruby, 2012; Tilman & Clark, 2014; Westhoek et al., 2014). For example, it has been argued that animal-sourced products contribute to biodiversity loss, climate change, soil loss, and water and nutrient pollution (Feeley, Machovina, & Ripple, 2015). Moreover, livestock and their byproducts are responsible for 51% of global greenhouse gas emissions (Goodland & Anhang, 2009), use one-third of the world’s fresh water, and produce an enormous amount of waste (Herrero et al., 2013). Additionally, livestock occupies about one-third of the world’s land and

contributes heavily to desertification (e.g. Herrero et al., 2013; Oppenlander, 2013). Deforestation and species extinction are two other disastrous by-products of animal agriculture, not to mention the ill treatment of animals in the livestock industry (Margulis, 2003; Oppenlander, 2013). Besides the enormous negative impact on the environment and animal welfare, animal products have been linked to health issues as well. The consumption of livestock products, especially red meat and processed meat, increases the risk for several illnesses such as cardio vascular disease, numerous types of cancer, and type 2 diabetes (Friel et al., 2009; Health Council of the Netherlands, 2011; McMichael et al., 2007).

The consequences of the consumption of meat and other livestock products are becoming more evident, making the required changes in consumer behavior more urgent as well (Palmer, 2010: 227). As stated by Roberts (2009: 209), the vast majority of forecasts indicate that lowering meat consumption is essential if we want to meet future food supply in a sustainable way. Influential documentaries, such as Cowspiracy by Andersen and Kuhn (2014), are stirring the public debate and are making consumers more aware of the urgent need to change food consumption patterns. Furthermore, scholars are interested in obtaining knowledge on consumers’ willingness to adopt a more plant-based diet and reducing meat consumption (e.g. Graça, Oliveira, & Calheiros, 2015). However, meat is still central to lots of people’s diet and one-fourth of Dutch consumers are considered “meat lovers”, which implies that they eat meat almost on a daily basis (de Bakker & Dagevos, 2012). Alarmingly, projections indicate that the global demand for livestock products will grow by 70% by 2050 (Gerber et al., 2013). It is therefore essential to find ways to reduce consumers’ meat intake. Consumers should be considered partners in the change process and thus need to be involved in realizing the shift to more sustainable forms of food consumption (de Bakker & Dagevos, 2012).

(9)

3

been conducted on the topic of organic food in the marketing field, but to the author’s current knowledge, very few studies in marketing have focused on plant-based food, let alone

preferences for meat and meat substitutes. While agricultural and nutrition journals

accumulated a lot of research on this topic so far, marketing has lagged behind. Besides, many of these studies are qualitative in nature or used other methods than conjoint analysis (e.g. multiple correspondence analysis in Graça, Oliveira, & Calheiros, 2015), and many have focused on understanding the attitude-behavior gap instead of reducing it (e.g. de Bakker & Dagevos, 2012; Austgulen, 2013; Vermeir & Verbeke, 2006). This research intends to advance marketing research by applying conjoint analysis to a rather uncharted topic in marketing. Choice-based conjoint analysis has received widespread acceptance in marketing research as well as marketing practice for measuring consumer preferences (Hennig-Thurau et al., 2007; Louviere & Woodworth, 1983).

Besides being valuable for academia, addressing the current research gap in the literature can also be beneficial for marketers. Although most companies focus on profits instead of maximizing social (including environmental) value (Prothero & Fitchett, 2000), it has been argued that firms can improve their overall performance by focusing on social value as well (Porter & Kramer, 2006). Green marketing, defined by Peattie (2001) as “marketing activities which attempt to reduce the negative social and environmental impact of existing products and production systems, and which promote less damaging products and services” (p. 129), is a topic that recently has gained more attention again (Wymer & Polonsky, 2015). Increasingly, firms become aware of the positive effects that environmentally friendly marketing strategies can have on the bottom-line (e.g. Luo & Bhattacharya, 2006). Thus, marketers can benefit from having knowledge about the preferences for meat substitutes. Understanding what customers find important in a meat substitute and what might motivate them to choose a meat substitute over conventional meat can help marketers in making their products more appealing to these customers, allowing firms to positively contribute to the environment and the bottom-line at the same time.

The following problem statement is being addressed in this research: How can consumers be motivated to replace meat by meat substitutes? To shed light on this problem, two central research questions will be answered. In order to understand consumers’

preferences, it is important to discover what product attributes they find important. The importance attached to the different attributes of the product will depend on different

consumer motives. Van Doorn and Verhoef (2011) frame the choice for an organic food as a social dilemma, meaning that consumers need to weigh several motives. They distinguish between individual motives and other-oriented motives (collective or social interests). This distinction is also considered relevant for this research, given that the choice for meat or meat substitutes is associated with a trade-off between individual motives such as hedonic desires and other-oriented motives, such as the impact on the environment. In turn, a trade-off between these motives will determine what attributes are playing an important role in

(10)

4

1. What attributes are playing an important role in consumers’ preferences for meat and meat substitutes?

Although the first research question will provide insight into consumers’ preferences, many previous studies suggest consumers often have the motivation and thus the intention to behave sustainably, but have a difficult time converting this into sustainable food choices and

consumption (e.g., Bray et al., 2011; Chatzidakis et al., 2007; de Boer et al., 2009; de Barcellos et al., 2011; Krystallis et al. 2009). The theory of planned behavior argues that behavior can be best predicted by the intention (i.e. motivation) the person has towards it (Ajzen, 1985). In turn, intentions are a function of the attitude (favorable or unfavorable evaluation) towards the behavior, subjective norm (perceived social pressure to perform or not to perform the behavior), and perceived behavioral control (the perception of the ease or difficulty of performing the behavior). However, there is often a gap between intention and behavior, the so-called intention-behavior gap. Thus, behavioral intentions do not reliably lead to changes in behavior (Sheeran, 2002; Armitage & Conner, 1999).

Intentions might differ from behavior for several reasons. One reason for this could be that the purchase of food, especially sustainable food, is continuously associated with trade-offs (i.e. there is a social dilemma – see van Doorn & Verhoef, 2011). Different product attributes compete with each other for consumer awareness, perceived relevance and influence on choice behavior (Grunert, Hieke, & Wills, 2014). Furthermore, intentions may differ from behavior because of the social desirability bias (Chang, Lusk, & Norwood, 2009). The perceived social pressure (subjective norm) or the tendency to give socially desirable answers is likely to predict intentions (stated preferences) but overstate true preferences. For instance, since it is socially desirable to care for the environment, respondents may state that they prefer a meat substitute to meat. Though, in reality, their true preferences may differ.

Since one can regularly observe a gap between intentions and behavior, especially in the context of sustainable behavior (de Bakker & Dagevos, 2012), this study applies

incentive- alignment. Incentive-alignment is an essential part of this research, since

consumers’ preferences for meat substitutes are likely to be overstated because of the social desirability bias. Therefore, it is likely that in reality the social dilemma is skewed more towards individual motivations instead of social/collective motivations. Incentive-alignment aims at motivating respondents to provide their “true” choices by for example promising to reward them with the alternative they chose in one randomly selected choice task. In this research, incentive-alignment will be used in an attempt to reduce the intention-behavior gap that is associated with the social dilemma and to observe how preferences change as a result. This is a valuable addition to marketing research, since most conjoint studies are based on hypothetical survey data and might therefore overstate preferences (Ding, Grewal, & Liechty, 2005). The second research question that will be addressed is:

2. How do consumers’ preferences for meat and meat substitutes change when incentive-alignment is applied?

(11)

5

aligned choice-based conjoint analysis. More specifically, this study intends to find out what will make meat substitutes more appealing to customers than meat and aims to highlight different consumer segments, which will need to be targeted differently as a result.

Subsequently, the products, packaging (e.g. nutritional labeling), and promotion efforts can be tailored to these specific consumer segments. It will contribute to marketing research by utilizing conjoint analysis for a relatively unexplored topic in marketing, and incorporating incentive-alignment to reduce any potential biases. As a result, this study will help advance marketing management as well as marketing research.

The remainder of this thesis is structured as follows. In the next chapter, the theoretical framework and hypotheses are outlined, which form the basis of the conceptual model.

Subsequently, the methodology is described and an analysis of the results is provided. Finally, this research concludes with a discussion, including managerial implications and limitations and directions for future research, and ends with a conclusion.

2 Theoretical Framework

This chapter will present a review of relevant literature in the field. First of all, the social dilemma that consumers are facing will be explained. Subsequently, the attributes considered relevant for this research are outlined and hypotheses are stated. Furthermore, the role of incentive-alignment as a moderator is discussed. Finally, the conceptual model is presented. 2.1 The social dilemma

The choice for one of the alternatives in this study can be framed as a social dilemma (van Doorn & Verhoef, 2011). Consumers will need to weigh individual-oriented motives such as health concerns and other-oriented motives (collective or social interests) such as

environmental concerns when choosing either a meat product or a meat substitute, and these motives are often in conflict with each other.

(12)

6

2.2 Type of product

2.2.1 Meat and meat substitutes

In Western food culture, meat still has a prominent position and is the center of meals

(Barrena & Sánchez, 2009; Douglas & Nicod, 1974; Holm & Møhl, 2000; Meiselman, 2000). Consuming meat is often associated with several individual-oriented motives. First of all, people can have hedonic motivations for consuming meat. According to Graça, Oliveira, and Calheiros (2015), meat is considered to be a source of pleasure for many people. More specifically, meat is particularly valued for its sensory properties, its unique taste, and texture (Grunert, Bredahl, & Brunsø, 2004; Issanchou, 1996). Besides hedonic motivations,

consumers can also possess health motivations. Consumers are increasingly concerned about their health. A growing number of people worldwide, especially in affluent Western countries, are facing increased levels of obesity. Distressed by growing obesity rates coupled with discussions about food safety, many consumers want to consume healthier food (Food

MarketWatch, 2008). Thus, not surprisingly, an often-cited personal motive for reducing meat consumption has been health concerns (Rozin et al., 1997). Thus, reducing meat consumption can be an essential part of following a healthy diet (Allen et al., 2000).

However, although meat can increase the risk for several diseases, it still has a strong image of being healthy and nutritional (de Bakker & Dagevos, 2012). Results of a study conducted by de Bakker and Dagevos (2010) show that the image of meat as being healthy functions as a barrier to consumers in moderating their meat consumption. More specifically, Dutch consumers seem to have their doubts about the healthiness of consuming less meat, and are uncertain whether a diet in which less meat is consumed is “balanced” (de Bakker & Dagevos, 2010: 139–141). Moreover, the majority of Dutch consumers are consuming more animal proteins than necessary and they remain unaware of it (PBL, 2009; Sebek & Temme, 2009). Thus, consumers may be under the impression that reducing meat consumption and replacing it with meat substitutes does not have a positive impact on their health when doing this a couple of times a week. Therefore, the majority of people may prefer meat to meat substitutes because they believe it is part of a healthy diet.

(13)

7

It is expected that consumers prefer meat to meat substitutes because it is a great source of pleasure and enjoyment and it is widely believed that meat is part of a healthy and nutritious diet. Even as many consumers seem to care about the environment and animal welfare, individual-oriented motivations might be stronger. Thus, it is highly likely that the majority of consumers prefer meat to meat substitutes because of strong hedonic desires as well as the belief that meat is healthy. Therefore, the following hypothesis is stated:

H1: Consumers prefer meat to meat substitutes 2.2.2 Grain-fed versus grass-fed

Animals can either be raised on a grain- or grass-diet. Grass-fed cows consume predominantly grass, while grain-fed cows are raised on a more unnatural diet of corn and soy (Gunnars, 2013). While this distinction is already very common in countries such as the United States (e.g. Umberger, Boxall, & Lacy, 2009), it is much less known in the Netherlands. While there are some companies in the Netherlands that sell grass-fed beef (e.g. Schotsehooglanders.nl and Puregrazefoods.nl), the options are scarce and mostly limited to home-delivery, meaning that the products are not available in stores. Although the supermarket chain Albert Heijn is selling “Irish Beef” from the brand Greenfields, the cows are not grass-fed all-year round and the packaging does not indicate clearly that it is a grass-fed product (Albert Heijn,

dierenwelzijn rund/kalf).

US consumers have indicated that the reasons for purchasing grass-fed cattle vary and are mainly based on health concerns, higher animal welfare, and environmental sustainability (van Elswyk & McNeill, 2014). Grass-fed beef often has a lower overall fat-content and is a much better source of Omega-3 fatty acids and CLA than grain-fed beef (van Elswyk & McNeill, 2014; Gunnars, 2013). Furthermore, it also contains more vitamin E and minerals, although grain-fed beef is also claimed to be highly nutritious (Gunnars, 2013). Besides, animal welfare standards are much higher for the grass-fed cow since it lives in a natural habitat and is not fed other supplements, hormones and antibiotics. Moreover, grass-fed beef is better for the environment since, among other reasons, less energy is needed for growing grass than grain. Additionally, farming grain-fed livestock affects the world food supply. As claimed by Pimental (1997), more than 800 million people could be fed in the United States with the grain that is currently being fed to livestock.

Several scholars have studied the importance of the grazing system in meat choices outside of the American context (e.g. Italian and Norwegian consumers in Hersleth et al., 2012; Spanish, French, and British consumers in Realini et al., 2013), but little is known about the effect of animal feeding on Dutch consumers’ preferences. Since the somewhat limited offering of grass-fed beef, Dutch consumers may not be aware of all the differences between the two types of feeding yet. Nevertheless, it is interesting to see whether the grain-fed vs. grass-grain-fed beef distinction plays a role in consumers’ decision process when presented with the information. Furthermore, it is worthwhile to discover whether it would be beneficial to offer more grass-fed beef and/or promote the distinction between the two types of feeding better to somewhat mitigate the negative effects associated with the consumption of meat, given that meat substitutes might be too far out of reach for certain consumers.

(14)

8

grain-fed beef because of health-, environmental-, and animal welfare motivations. In this case, hedonic motivations do not lead to preferring one option to the other since both types of beef can be a source of pleasure and enjoyment. Thus, it is expected that consumers prefer grass-fed to grain-fed beef because of health-, environmental-, and animal welfare concerns. H1a: Consumers prefer grass-fed beef to grain-fed beef

2.2.3 Basis of meat substitutes: soy/lupine or vegetables

Most meat substitutes are based on legumes (such as soy, lentils, (chick) peas or lupines), wheat, rice, or egg protein (Broekema & Smale, 2011). The meat substitutes on the Dutch market that are made from soy or lupine often closer resemble the flavor of meat than meat substitutes that are based on vegetables (e.g. “veggie” burgers, which often include a variety of vegetables such as corn, onions, peppers, and carrots). De Vegetarische Slager (The Vegetarian Butcher) is a company launched in 2010 in the Netherlands, which has succeeded to almost completely mimic the taste of real meat by using soy and lupine as the main

ingredients of their products (Devegetarischeslager.nl). Their products include substitutes for chicken, meatballs, hamburgers, bacon, and tuna among others.

Although altering consumption patterns is challenging, there is promising research indicating that vegan (non-animal) foods are often similarly evaluated as animal-based foods when they resemble each other. According to the law of similarity, when things look alike, they are perceived as having the same properties or essence (Adise, Gavdanovich, & Zellner, 2015). The law of similarity implies that meat substitutes can evoke the same reactions as meat does. As argued by Hoek, van Boekel, Voordouw, and Luning (2011), this may especially be the case for meat substitutes that are processed (e.g. meatballs and burgers), because these are considered to be from the same food category as their meat-counterparts are. The success of a company such as the Vegetarian Butcher confirms the fact that meat substitutes need to resemble meat in order for them to be regarded as replacements by

consumers (Hoek et al., 2011). Thus, meat substitutes that resemble meat in their appearance as well as flavor are a promising product given that reactions to meat are positive. Therefore, from the perspective of hedonic motivations, it is expected that consumers would prefer soy/lupine-based meat substitutes to vegetable-based ones, since they would still be able to enjoy “meat”.

Although both soy/lupine and vegetables are healthy, consumers may prefer

(15)

9

percentage of the soybeans are fed to livestock. More specifically, approximately 75% of soybeans are used for animal feed, placing the environmental burden mainly on the livestock industry and the consumers that consume soy indirectly through the consumption of animal-products (WWF Global, soy facts & data). Thus, the environmental impact of people

consuming soy-based meat substitutes is much lower. However, it could become a problem if a large number of people are substituting meat for soy-based meat substitutes in the future.

To conclude, it is likely that consumers will prefer soy/lupine-based meat substitutes to vegetable-based meat substitutes because of hedonic desires (i.e. the product is more similar to meat) and health motivations (mainly related to protein), as well as the negligible negative environmental impact compared to vegetables it currently has. Therefore, the following hypothesis can be stated:

H1b: Consumers prefer soy and/or lupine-based meat substitutes to vegetable-based meat substitutes

2.3 Production method

Consumers cannot shop at a supermarket anymore without noticing the increasing number of organic products. In the Netherlands, the share of organic food is growing every year,

although the growth rate decreased slightly in 2014 (Monitor Duurzaam Voedsel, 2013). The purchase of organic food has frequently been associated with personal health motivations (e.g. Baker et al., 2004; Lusk & Briggeman, 2009, Pino et al., 2012; Gottschalk & Leistner, 2013). Hughner et al. (2007) found that health concerns are among the top 5 purchase motives for organic food. Undeniably, the majority of consumer studies continue to show that

expectations concerning health effects of organic food are among the most important reasons for consumers to buy organic products. Even though many of these expectations regarding health benefits lack thorough scientific evidence (if health effects are defined as effects on defined human diseases), there is still a widespread belief among consumers that organic food is healthier than conventional food (Huber et al., 2011).

Furthermore, consumers can be motivated by environmental concerns (e.g. Kollmuss & Agyeman, 2002; Hokanen et al., 2006; Brown et al., 2009; Pino et al., 2012; Gottschalk & Leistner, 2013). Since organic farming uses less pesticides and artificial fertilizers, it is believed to be less harmful to the environment (Cornelissen et al., 2008; Gore, 2006; The Week, 2009). Furthermore, animal welfare concerns could play a role as well. Animal welfare can be defined in many ways (Spoolder, 2007). In the Netherlands, livestock raised in an organic farming system have more space to move, are outside more often, consume organic food, and receive antibiotics only when they are sick (Het Voedingscentrum). Finally, the choice between organic and non-organic can be motivated by hedonic motivations also (e.g. Brown et al., 2009). More specifically, organic food is often associated with better quality and a better taste as well (Gracia et al., 2014).

Therefore, based on personal health motivations, environmental- and animal welfare motivations, and hedonic desires, it is likely that consumers will prefer organic to

(16)

10 H2: Consumers prefer organic to non-organic

2.4 Nutritional claims

As mentioned before, people are becoming increasingly concerned about their health. In general, marketers tend to promote healthy food in a cognitive way, meaning that marketing communication often focuses on the nutrition facts as well as the health benefits associated with the product (Hastings et al., 2003). Numerous previous studies have suggested that healthy food choices can be encouraged by optimizing nutrition labeling on food products (Grunert & Wills, 2007; Nayga, 2008; Verbeke, 2008). In fact, nutritional claims and health-related product messages have been shown to be successful in many food categories (Bublitz & Peracchio, forthcoming). According to Mittelstaedt, Kilbourne, and Shultz (forthcoming), promoting healthy choices can enhance the consumers’ health while at the same grow the bottom-line of businesses that market these healthy alternatives.

Meat often contains a high amount of saturated fat, which increases blood cholesterol, and consuming this frequently is considered bad for your health (The Heart Foundation of Australia). Though, consumers also have the option to purchase low-fat meat. Koistinen et al. (2013) showed the importance attached to low fat and found that low fat has a positive effect on the choice of a minced meat product.Furthermore, as discussed before, many consumers are still under the impression that meat is the best source of protein, and it is still the main source of protein in high-income countries while many other alternative protein sources exist (Grigg, 1999). Since consumers may not be aware of the fact that meat substitutes can be high in protein as well, it will be interesting to see how this nutritional claim influences consumer preferences.

Therefore, it will be interesting to discover what the effects of several different nutritional claims regarding fat and protein are on consumers’ preferences for meat and meat substitutes. While some studies have shown that health claims are not the best way to

convince consumers that are attached to meat (e.g. to its flavor) to reduce their meat consumption (Hoek et al., 2011), it might work better for consumers that already consider eating less meat or can be characterized as flexible meat eaters (de Backer & Hudders, 2014). However, since in general consumers increasingly care about their health, it is likely that consumers prefer a product with a nutritional claim to a product that has no such claim. Therefore, the following is expected:

H3: Consumers prefer high protein and low-fat to no nutritional claims 2.5 Incentive-alignment

Choice-based conjoint analysis is a popular method to measure consumer preferences. The majority of conjoint studies are hypothetical (i.e. a hypothetical bias exists), implying that there are no real consumption consequences for people participating in the study. When there are no incentives for respondents to provide truthful answers, they may experience a lack of involvement (e.g., Ding et al., 2005; Netzer et al., 2008). It is highly probable that respondents exert less effort and attention when filling out the preference measurement survey than when they would be making real purchase decisions (Toubia, de Jong, Stieger, & Füller, 2012). This implies that stated preferences may differ from true preferences. As a result of the

(17)

11 2004). Therefore, the extent to which conjoint analysis is measuring real consumer

preferences is debatable (Ding et al., 2005).

Incentive-alignment mechanisms that create a link between responses and a potential reward can encourage participants to answer more truthfully (Ding et al., 2005; Ding, 2007). There are many different incentive-alignment mechanisms that can reduce hypothetical bias. For example, the reward may be selected from the chosen alternatives in the experiment (Ding et al., 2005). Research has shown that the external validity of choice-based conjoint analysis is much higher when responses influence compensation (e.g. Ding et al., 2005). Incentive-alignment mechanisms have been found to reduce hypothetical bias and increase predictive validity as well (e.g. Ding et al., 2005; Ding, 2007; Dong, Ding, & Huber, 2010; Miller et al., 2011). Since it is likely that an intention-behavior gap exists in the context of this research, incentive-alignment can help elicit true preferences and can make stated preferences better predictors of actual purchase behavior.

One bias that contributes to the hypothetical bias is the social desirability bias. In an experimental setting, people know that their behavior will be observed which may lead them to provide socially acceptable answers (Chang, Lusk, & Norwood, 2009). The notion that stated preferences may differ from true preferences because of the social desirability bias is especially relevant in the context of this research, since respondents are facing a social dilemma, meaning that they have to make a trade-off between individual-oriented and social-oriented motivations. Most consumers are aware of the fact that organic and plant-based meat is better for the environment than conventional meat is. Although consumers may personally prefer conventional meat, they may state that they prefer organic meat because this is a socially desirable answer. Therefore, the preferences for organic meat and meat substitutes might be overstated. Overall, since it is socially desirable to care about your own health as well as the environment and animal welfare, it is likely that these motivations are playing a weaker role when the experiment is incentive-aligned. When respondents are encouraged to provide their true preferences by means of incentive-alignment, it is likely that hedonic motives (or other individual motives not discussed in this research, such as convenience) or other factors such as price (which will be included as a control variable) become more powerful than health concerns and other-oriented motivations. Therefore, the following is expected:

H4a: When using incentive-alignment, consumers place more value on hedonic motives, thus decreasing the effect of meat substitutes on utility compared to meat

H4b: When using incentive-alignment, consumers place more value on hedonic motives, thus decreasing the effect of vegetable-based meat substitutes on utility compared to soy/lupine-based meat substitutes

H4c: When using incentive-alignment, consumers place more value on hedonic motives, thus decreasing the effect of organic on utility compared to non-organic

(18)

12 2.6 Conceptual model

Figure 1 below displays the conceptual model, which is derived from the hypotheses.

Figure 1: Conceptual model

3 Methodology

In this section, the method that is used to collect insights is explained. Then, the procedure and sample are described. This is followed by an explanation of the specific type of incentive-alignment used. Then, the study design and experimental design are outlined, including the corresponding attributes and levels. Finally, the scales used in the final part of the survey are explained and the model is specified.

3.1 Method

This study uses preference measurement in order to test the hypotheses and subsequently answer the proposed research questions. Preference measurement is among the most important topics in marketing research and can reveal what customers like and prefer. Additionally, preference measurement can give insight into consumers’ underlying motives. To elicit consumer preferences, this study uses a decompositional approach, meaning that respondents have to evaluate entire products by considering the product attributes and levels jointly (i.e., conjoint analysis). These stated preferences can be decomposed with statistical methods to obtain part worth functions (Eggers & Sattler, 2011; Rao, 2014).

(19)

13 purchase situations (Toubia, Hauser, & Simester, 2004). CBC analysis is said to be more effective because consumers are making many choices in their daily life. Thus, choices result in more “natural” manifestations of consumers’ preferences (Eggers & Sattler, 2011). As a result, it has been assumed that CBC analysis gives more accurate results (Toubia, Hauser, & Simester, 2004).

3.2 Procedure and sample

The survey is structured as follows: an introductory page with a brief explanation of the study and the possibility to win a restaurant coupon, which differs per condition as explained in the next section; an explanation page about the different types of hamburgers that a restaurant offers; a description of the scenario; 13 choice sets; the health concern scale; the New Ecological Paradigm Scale (NEPS); two hedonic motives scales; and demographic information including gender, age, and highest completed educational level.

In order to collect the data, the survey is distributed through the following channels: e-mail, Facebook (relevant groups), and LinkedIn. There are several restrictions on

participation. First of all, only Dutch people are able to participate, and vegetarians are excluded from participation since all the alternatives would not be equally attractive to them (i.e. they would never choose meat). Respondents receive a link to the survey in Preference Lab, which is a tool that is well equipped for CBC analysis. Although there are no specific requirements regarding the demographic profile of respondents, the survey also collects the age, gender and highest completed educational level of the respondents in order to describe the sample better afterwards.

3.3 Incentive-alignment

In order to test the moderating effect of incentive-alignment, two different experimental conditions are developed, namely a hypothetical choice-based conjoint condition and an incentive-aligned choice-based conjoint condition. Participants are allocated randomly to one of the two conditions. Participants in the hypothetical condition are not bound by their

responses, whereas participants in the incentive-aligned condition face consequences and have to “live” with their choices (e.g. see Ding et al., 2005). A between-subjects design is

employed, meaning that each respondent is put randomly in one of the two conditions. The specific incentive-alignment mechanism that is used in this research is the RankOrderIA mechanism (see Dong, Ding, & Huber, 2010). Before the conjoint survey, participants are told that they can win a restaurant coupon worth 15 euros (per 50

respondents), which can be used in a particular restaurant in the respondent’s area.

(20)

14 3.4 Study design

The first step of designing a conjoint study is to decompose the product into attributes and levels. In this research, the products of interest are hamburgers. This specific product is chosen because it is one of the meat products on the market that comes in a variety of non-animal based forms as well. Thus, the majority of people are likely to be familiar with the existence of the non-animal hamburgers as well. Additionally, as mentioned before, especially meat substitutes that are processed can evoke the same reactions as meat does (Hoek et al., 2011).

In general, it has been stated that a study design should only include a few attributes, usually no more than six, because it would become increasingly complex for respondents to fill out the survey (Eggers & Sattler, 2011). Therefore, five different attributes have been selected for this research and form the basis of the CBC analysis, i.e. type of product, animal feed, basis of product, production method, and nutritional claim. The attributes animal feed and basis of product are conditional on the type of product, i.e. when the type of product is meat, the attribute animal feed will be displayed. Similarly, when the type of product equals meat substitute, the attribute basis of product will be displayed. Therefore, it was decided to combine the attributes, type of product, animal feed, and basis of product for the purposes of the questionnaire. More specifically, the attributes in the survey are displayed as follows: type of product (grass-fed beef, grain-fed beef, based on soy and/or lupine, based on vegetables), production method (organic, non-organic), and nutritional claim (high protein and low fat, high protein, low fat, no claim).

Besides these attributes of interest, one control variable is included as an additional attribute in order to make the choice situation more realistic. Price is almost always an important factor that influences consumers’ preferences for products. In conjoint analysis, respondents often need to make a trade-off between price and other relevant attributes. Van Doorn and Verhoef (2011) have shown that consumers may not be willing to pay the price premium that is associated with an organic product. Meat substitutes are often more expensive than their regular meat counterparts as well, as is the case with most “green” products (Olson, 2013). Although price is not the main focus of this research, price is included as a control variable in order to make the choices more realistic. The attribute price contains four levels, namely €9.50, €10.50, €11.50, and €12.50. Table 1 gives an overview of the attributes and levels included in the choice design.

Table 1: Attributes and levels

Attributes Levels

Primary attribute

Type of product § Meat

§ Meat substitute Conditional attribute Animal feed Conditional attribute Basis of product § Grass-fed § Grain-fed

§ Based on soy and/or lupine § Based on vegetables

Common attribute

Production method § Organic

(21)

15

Common attribute

Nutritional claim § High protein and low-fat

§ High protein § Low-fat § No claim Common attribute Price § €9.50 § €10.50 § €11.50 § €12.50 3.5 Experimental design

The experimental design refers to the attribute-level combinations shown to respondents. Since the experimental design has a direct effect on the reliability of results, it is a particularly important part of CBC analysis (Eggers & Sattler, 2011). Whereas a full-factorial design would show all possible attribute-level combinations, a fractional factorial design shows only a subset of the full factorial. In order to determine the optimal efficient choice design, Huber and Zwerina (1996) developed several criteria. First of all, the choice design should be balanced, which implies that all the attribute levels have to be displayed an equal number of times. Secondly, all the attribute-level combinations should be shown an equal number of times, which is also referred to as orthogonal design. Thirdly, there should be minimal overlap between the choice sets, meaning that the alternatives should be different from each other and attributes should not have the same levels within one choice set. Finally, the choice sets should be non-dominated, which implies that all alternatives within a choice-set should be equally appealing to respondents.

A 4x2x4x4 experimental design is employed. Since a full factorial design is

unfeasible, participants have to answer a subset of the full factorial. The stimuli are randomly assigned to the different choice sets. In total, respondents are presented with 13 choice sets. To account for minimal overlap and to reduce the complexity of the decisions, every choice set consists of two alternatives. The survey captures one dependent variable, namely the best choice. In the majority of experiments, a no-choice option is included to make the experiment more realistic. Although this option can increase the resemblance to real-life purchase

situations, there is no real obligation for participants to buy their chosen option, meaning that the no-choice decisions may be underrepresented (Wlömert & Eggers, 2014). Furthermore, data is lost when participants select the no-choice option. Therefore, it is decided to include a dual response no-choice option instead of a no-choice option. After selecting their preferred choice, participants are asked to indicate whether they would actually order their preferred alternative.

(22)

16

Figure 2: Exemplary choice set

3.6 Scales

In order to explain the results further, several control variables are included by means of scales that capture the respondents’ health concern, environmental concern, and hedonic motives. The health concern scale is adopted from Kähkönen, Tuorila, and Rita (1996)and includes 10 items that are measured on a 7-point Likert scale (from completely disagree to completely agree). The instrument used for measuring environmental concern is the New Ecological Paradigm scale (NEPS), originally developed by Dunlap and van Liere (1978). The scale was adapted in 2000 by the same authors and consists of 15 items, which measure general beliefs about the relationship of human beings to the environment. Respondents are asked to indicate to what extent they agree with those statements on a 7-point Likert scale. Finally, two types of scales are used to measure respondents’ hedonic motives. The first scale consists of five items and was developed by Olsen, Thach, and Hemphill (2012). It measures values related to a hedonistic life, making it a more abstract measure of hedonic motives. The second scale consists of four items and was developed by Graca, Oliveira, and Calheiros (2015). This specific scale is chosen because it refers to meat as a source of pleasure, which is more specific to this research. An overview of the scales can be found in Appendix 7.1. 3.7 Model

In the random utility model, choices are based on the overall utilities of the alternatives. The utility U of consumer i for hamburger j is a latent construct and is specified by the formula below, where V represents the systematic utility component (rational utility) and ε the stochastic utility component (error term), of which the latter captures the effects that are not systematic and thus not accounted for (Manski, 1977).

𝑈_!" = 𝑉_!" + 𝜀_!" (1)

(23)

17 specifically, the hamburgers in this case are a combination of type of product, production method, nutritional claim, and price. It is assumed that consumers choose the alternative with the highest utility. The systematic utility of consumer i for hamburger j, which is the sum of the part-worth utilities attached to each attribute, is shown below:

𝑉_!" = 𝛽_!𝑇𝑂𝑃1_!+ 𝛽_!𝑇𝑂𝑃2_!+ 𝛽_!𝑇𝑂𝑃3_!+ 𝛽_!𝑃𝑀1_!+ 𝛽_!𝑁𝐶1_! + 𝛽_!𝑁𝐶2_!+

𝛽!𝑁𝐶3!+ 𝛽!𝑃𝑟𝑖𝑐𝑒! (2)

Where:

Vij = systematic utility of consumer i for hamburger j

β1...β8 = part-worth utilities

TOP1j = effect-coded variable for grass-fed beef for hamburger j

TOP2j = effect-coded variable for grain-fed beef for hamburger j

TOP3j = effect-coded variable for soy/lupine-based meat substitute for hamburger j

PM1j = effect-coded variable for organic for hamburger j

NC1j = effect-coded variable for high protein and low fat for hamburger j

NC2j = effect-coded variable for high protein for hamburger j

NC3j = effect-coded variable for low fat for hamburger j

Pricej = linear variable for price in euros for hamburger j

To account for the moderation effect of incentive-alignment, interaction terms are added to the model in the form of the multiplied effects of incentive-alignment (which is 1 or 0, since the consumer can either be in the IA or hypothetical condition) with the effect codes of the different attributes of interest. The mathematical equation is shown below.

𝑉_!" = 𝛽_!𝑇𝑂𝑃1_! + 𝛽_!𝑇𝑂𝑃2_!+ 𝛽_!𝑇𝑂𝑃3_!+ 𝛽_!𝑃𝑀1_!+ 𝛽_!𝑁𝐶1_!+ 𝛽_!𝑁𝐶2_!+ 𝛽_!𝑁𝐶3_!+ 𝛽_!𝑃𝑟𝑖𝑐𝑒_!+ 𝛽_!(𝐼𝐴_! ∗ 𝑇𝑂𝑃1_!) + 𝛽_!"(𝐼𝐴_! ∗ 𝑇𝑂𝑃2_!) + 𝛽_!!(𝐼𝐴_! ∗ 𝑇𝑂𝑃3_!) + 𝛽_!"(𝐼𝐴_!∗ 𝑃𝑀1_!) + 𝛽_!"(𝐼𝐴_! ∗ 𝑁𝐶1_!) + 𝛽_!"(𝐼𝐴_!∗ 𝑁𝐶2_!) + 𝛽_!"(𝐼𝐴_! ∗ 𝑁𝐶3_!) (3) The systematic utility of the alternatives can be estimated with the multinomial logit (MNL) model (e.g. McFadden, 1974; Islam, Louviere, & Burke, 2007; Louviere, Hensher, & Swait, 2000). The choice of hamburger j from choice set C is a function of the overall utility. The probability that an alternative will be selected from a given choice set C is captured in this MNL choice model. Thus, modeling the utility of different hamburgers allows a prediction of the choice probabilities. The MNL model is shown below.

𝑃𝑟𝑜𝑏 𝑗 𝐶 = !"#(!!) !"#(!!) !

!!! (4)

(24)

18

4 Results

This section discusses the findings of the online conjoint survey. First of all, a description of the sample’s demographic profile is given. Secondly, the results of factor analyses and reliability analyses regarding the scales are outlined. Thirdly, the model fit is assessed, which is followed by a general analysis. Fourthly, the hypotheses are tested and the results are summarized. Finally, a segmentation analysis is carried out.

4.1 Sample

The survey was completed by 92 respondents, from which 26 are male and 66 are female. More specifically, 41 people participated in the hypothetical experiment, whereas 51 people participated in the incentive-aligned experiment. Figure 3 shows the distribution of age and the highest completed degree among respondents. As can be seen in figure 3, the majority of respondents are between 18 and 29 years old, and about 60% of them either obtained an HBO or a University degree.

Figure 3: Descriptives

In order to check whether the two conditions are similar to each other regarding the

demographic characteristics, several tests were performed. Chi-square tests were performed to test whether the two conditions can be regarded as structurally similar regarding gender, age, and degree. More specifically, a cross-tabulation was performed for IA (0 and 1) and gender, age, and degree. The Pearson Chi-Square statistic for IA*age is 229,987 (p=0,000). Thus, the two conditions are not entirely similar regarding age. More specifically, the IA condition consists mainly of 18-29 year olds, and for about 11% of 50-59 year olds. The no IA also consists mainly of 18-29 year olds, but has more 40-49 year olds (10%) than the IA condition (2%). Regarding IA*gender, the Pearson Chi-Square statistic amounts to 1,925 (p=0,165). Therefore, there is no statistically significant association between the conditions and gender; both conditions consist of more females than males. The Pearson Chi-Square statistic for IA*degree amounts to 102,22 (p=0,000). So, the two conditions are not structurally similar

(25)

19 with regard to degree. Overall, the participants in the no IA condition are somewhat less highly educated. About 44% have not completed a degree higher than MBO, whereas that is 37% in the IA condition. Further inspection reveals that the biggest difference exists

regarding the HBO degree. More specifically, of the participants in the no IA condition and the IA condition, 19,5% and 27,5% obtained an HBO degree, respectively. Thus, it can be concluded that the conditions are structurally similar with regard to gender, but there are some differences with regard to age and degree. The IA condition consists of relatively more young people (18-29) and people in their fifties, and they are slightly higher educated in general. This is not ideal, but the differences are not very large and can therefore be assumed to have small effects on the results, if any.

4.2 Scale reliability

In order to discover whether the items (all measured on a 7-point Likert scale: 1=totally disagree, 7=totally agree)included in the different scales indeed capture some underlying construct, factor analysis as well as reliability analysis was conducted. Principal component analysis (PCA) is used to find common variance and retrieve underlying dimensions. More specifically, it is used to check whether the items capture one construct (e.g. health concern), or whether they are multidimensional and thus capture multiple constructs. The KMO

measure of sampling adequacy, Barlett’s test of sphericity and the communalities are used to determine whether performing factor analysis is appropriate. Then, it is decided whether the items can be captured by one factor only or need to be represented by more factors. Reliability analysis in the form of Cronbach’s alpha is done to assess the internal consistency, i.e. how closely related the items are as a group (or how reliable the scale is). Furthermore, the Cronbach’s alpha “if item deleted” is inspected as well. The resulting factors are used in an attempt to explain the results that are obtained by conjoint analysis. The details can be found in Appendix 7.2.

4.2.1 Health concern

The tests reveal that factor analysis is appropriate (see Appendix 7.2 for all details). The KMO measure of sampling adequacy amounts to 0,788 and Bartlett’s test of sphericity is significant. Thus, there is significant correlation between the items. Finally, the

communalities are not all higher than 0,4. More specifically, the fifth item, “Getting sufficient energy from my food”, has a communality of 0,334. Therefore, it was decided to delete this variable from the analysis. This improves the KMO to 0,794. Reliability analysis confirms this decision: the Cronbach’s alpha increases from 0,851 to 0,859 when item five is deleted. Thus, the final scale, which is highly internally consistent, has nine instead of ten items.

Next, the amount of factors is chosen. The Eigenvalues indicate that a three-factor solution is most appropriate. The total explained variance of these three factors is 81,3%, whereas using one factor would result in only explaining 48,42% of the common variance. Thus, continuing with one underlying construct is inappropriate. The nine items have to be captured by three factors instead of one, i.e. the scale is not unidimensional, but

(26)

20 lot of cholesterol in my food, factor two of getting a lot of salt, fat, sugar, and food additives in my food, and factor three of getting many calories and gaining weight. Thus, factor one captures heart-related concerns, factor two has to do with getting too much of an unhealthy substance, and factor three is related to weight-gains. The table below shows an overview of the factors and the corresponding items.

Table 2: Health concern factors

HC 1 HC 2 HC 3

7: Risk for high blood pressure 8: Risk for coronary heart disease 9: Getting a lot of cholesterol in my food

1: Getting a lot of salt in my food 2: Getting a lot of fat in my food 3: Getting a lot of sugar in my food 6: Food additives in my food

4: Getting many calories 10: Gaining weight

4.2.2 Environmental concern

Amburgey and Thoman (2012) found that treating the scale as a single factor reflecting environmental concern neglects the multiple dimensions that are present. In order to test whether this is true, confirmatory factor analysis (CFA) is performed and one factor is extracted. Although the KMO amounts to 0,66 and Barlett’s test of sphericity is significant, many of the communalities are below 0,4. Furthermore, having one component results in only explaining 25,84% of the common variance. Therefore, it is very unlikely that the scale is unidimensional. Besides, reliability analysis by means of Cronbach’s alpha gives a value of 0,727, but results in having to delete many items to increase Cronbach’s alpha.

Instead, Amburgey and Thoman (2012) suggest treating the scale as five interrelated facets by using confirmatory factor analysis (CFA) within a structural equation modeling approach, i.e. using a second-order facture structure. However, since this complicates the analysis of the results even further and because SPSS does not allow for this, five orthogonal factors were extracted separately, representing the items tested by Amburgey and Thoman (2012). However, this solution appears to be suboptimal – the conceptualized factor structure does not hold for this specific sample. For example, the three items specified do not best represent the dimension “balance of nature”, as originally conceptualized by Dunlap and van Liere (2000). Therefore, it was decided to perform PCA without specifying the number of factors extracted. The results indeed indicate that a five-factor solution is most appropriate. After rotating the component matrix with VARIMAX, item four has a factor loading below 0,4. Therefore, item four is deleted. This results in five factors that explain about 68% of common variance. However, even after factor rotation, a few cross-loadings can be observed (item eight and ten). Thus, for several items it is not entirely clear to which factor they belong. The final factor structure is specified in the table below.

Table 3: Environmental concern factors

EC 1 EC 2 EC 3 EC 4 EC 5

1: We are approaching the limit of the number of people the Earth can support.

3: When humans interfere with nature it often produces disastrous consequences.

12: Humans were meant to rule over the rest of nature. (-) 14: Humans will eventually learn enough about how

6: The Earth has plenty of natural resources if we just learn how to develop them. (-) 10: The so-called “ecological crisis”

9: Despite our special abilities, humans are still subject to the laws of nature.

2: Humans have the right to modify the natural environment to suit their needs. (-)

(27)

21

5: Humans are seriously abusing the environment.

7: Plants and animals have as much right as humans to exist. 11: The Earth is like a spaceship with very limited room and resources.

13: The balance of nature is very delicate and easily upset. 15: If things continue on their present course, we will soon experience a major ecological catastrophe.

nature works to be able to control it. (-)

facing humankind has been greatly

exaggerated. (-)

the impacts of modern industrial nations. (-)

(-) Items that were reverse-scored

4.2.3 Hedonic motives

The scale that was adopted from Olsen, Thach, and Hemphill (2012) has a KMO that amounts to 0,809 and the Barlett’s test of sphericity is significant, which indicates that factor analysis is appropriate. However, not all the communalities are above 0,4: the communality of the first item does not meet the threshold. Cronbach’s alpha also increases from 0,834 to 0,846 if the first item is deleted. Therefore, the first item was dropped from the analysis. Component one is the only factor that exceeds the Eigenvalue of one and explains 70,97% of the common variance. Therefore, there is evidence of a unidimensional scale. This implies that the items (two till five) can be captured by one factor.

The second scale, adopted from Graca, Oliveira, and Calheiros (2015), has a KMO that amounts to 0,751, Barlett’s test of sphericity is significant, and all the communalities are higher than 0,4. Thus, factor analysis is considered appropriate. Although the Cronbach’s alpha would increase from 0,88 to 0,892 if the last item (“A good steak is without

comparison”) were deleted, the communality for this item amounts to 0,62, which is sufficiently high. Therefore, it was decided to proceed with all the items. The Eigenvalues indicate a one-factor solution, and this component explains 75,23% of the common variance. Therefore, it can be concluded that the scale is unidimensional: the four items measure one underlying construct.

4.3 Model fit: goodness of fit comparison

First of all, the model specification for each attribute is determined. The type of product, production method, and nutritional claim are nominal variables, so they have to be specified as part-worth attributes by default. In order to determine whether there is a linear relationship between the attribute levels of price and the corresponding utility values, the utility

parameters are observed. Higher prices are less preferred, so price seems to have a negative linear relationship with utility (see figure 5 in the next section).

(28)

22 Chi-square statistic amounts to 0,9346, which is lower than the critical value (5,991).

Therefore, model one is not significantly better than model two. Furthermore, the R2 adjusted (which adjusts for the number of parameters) and hit rate are almost the same as well.

Therefore, it was decided to continue with the more parsimonious model: model two. Finally, the Log-likelihood of the model with parameters is compared to the Log-likelihood of the Null model (no parameters) by means of the Likelihood-ratio test. The Chi-square statistic amounts to 739,172with nine degrees of freedom, which gives a critical value of 23,589 (alpha is 5%). It can be concluded that the estimated model parameters are significantly different from zero (P-value Chi-square < 0,0001).

Table 4: Goodness of fit comparison

Goodness of fit comparison

Model 1 (price nominal): Model 2 (price linear):

LL(0) -1658,008 -1658,008 LL(β) -1287,9547 -1288,4220 Parameters: 11 9 R2 _adjusted _0,22 _0,22 Hit rate 73,62% 73,66% Prediction error 26,38% 26,34% 4.4 Preference estimates 4.4.1 Attribute importance

In order to measure how much influence each attribute has on respondents’ choices, the ranges within an attribute are calculated. The no-choice option is not included here. The attribute importance results can be found in figure 4 below. Product type is most important to consumers (52,27%), followed by price (22,07%), production method (17,05%), and

nutritional claim (8,61%). However, the importance that is attached to each of these attributes is measured on an aggregate level (i.e. for the average consumer). In other words, preference heterogeneity has not been taken into account. Furthermore, the moderating effect of

(29)

23

Figure 4: Attribute importance

4.4.2 Average choice shares compared to no-choice option

The survey that is used for this research includes a dual-response no-choice option. Therefore, the average choice shares of two realistic products are compared to the no-choice option, to determine the probability of choosing the alternative over the no-purchase option. The utility of the alternatives is compared with the utility of the no-purchase option, from which the probabilities are calculated. As can be seen in the table below, the probability that a grass-fed, organic, low-fat hamburger of €11,50 is chosen over the no-purchase option is 0,51. Thus, when a respondent chooses this hamburger in a certain choice set J, there is 51% chance that this respondent will choose “yes” when asked whether he/she would really order this specific hamburger at the restaurant. The probability of choosing a vegetable-based, non-organic, high protein and low fat hamburger of €10,50 over the no-purchase option is much lower, only 22%. Thus, when the hamburger has these specific attribute levels, it is much more likely that the respondent decides to not order this alternative he/she chose in a certain choice set J.

Table 5: Average choice shares compared to no-choice option

Alternatives Utility estimates Total utilities and probability

Grass-fed 0,7841 Alternative: -1,4734

Organic 0,2461 No-choice: -1,5141

Low-fat -0,0621 Probability alternative: 0,51

€11.50 -0,2123*€11.50 Probability no-choice: 0,49

Vegetable-based -0,4447 Alternative: -2,75895

Non organic -0,2461 No-choice: -1,5141

High protein and low fat 0,1610 Probability alternative: 0,22

€10.50 -0,2123*€10.50 Probability no-choice: 0,78 0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% Product

type Produc9on method Nutri9onal claim Price

A1ribute importance

(30)

24 4.5 Hypotheses testing

4.5.1 Main effects

Based on the model specified above, aggregate model two, the main effects are investigated. The utility parameters plotted below already give an indication of consumers’ preferences regarding the levels of the different attributes.The utility estimates indicate that meat-based hamburgers are preferred to meat substitutes. More specifically, consumers prefer grass-fed beef the most, followed by grain-fed beef, vegetable-based hamburgers, and soy/lupine-based hamburgers. Furthermore, organic is preferred to non-organic, nutritional claims to no claim, and low prices to high prices.

Figure 5: Model specifications and corresponding utility estimates

Table 6 below displays the utility estimates, Wald-values, and corresponding p-values.As can be derived from the table, the attributes type of product, production method and price have a significant effect on utility. Nutritional claim only has a marginally significant effect on utility (p=0,056). Also, price has a significant and negative effect on utility, which implies that an increase in price corresponds to a decrease in utility. Although this means that the effect of the attributes on utility is significantly different from zero, it does not say anything about the different levels yet. Therefore, in order to test H1-H3, it was investigated whether the levels of interest differ significantly from each other in their effect on utility.

In order to test whether H1 can be accepted, the difference between the two closest utility values, the least preferred meat type and the most preferred meat substitute type, is tested. The difference between the two utility estimates is found to be significant (Z-value 13,67 > 1,96). Therefore, H1 is accepted. As a result, it can also be concluded that consumers significantly prefer grass-fed beef to grain-fed beef (Z-value 6,51 > 1,96). Thus, H1a is accepted as well. As can be seen in table 6 below, consumers prefer vegetable-based hamburgers to soy/lupine-based hamburgers, which is contrary to expectations. It is even found that vegetable-based hamburgers are significantly preferred to soy/lupine-based hamburgers (Z-value -4,50 < -1,96). Therefore, H1b is rejected.

(31)

25 effect is tested and is found to be significant (Z-value 15,98 > 1,96). Therefore, H2 is

accepted. Finally, H3 states that consumers prefer high protein and low fat to no nutritional claims. As mentioned before, this attribute only has a marginally significant effect on utility. Further tests reveal that only the utility of high protein and low fat (level one) is significantly different from the utility of no nutritional claim (Z-value 4,05 > 1,96). High protein (level two) and low fat (level three) do not significantly differ from no nutritional claim (Z-value -1,24 and -0,21, respectively). Thus, since the effect of the attribute on utility is marginally significant, and because only the combined effect of high protein and low fat (level one) is significantly different from no nutritional claim, H3 is only partially supported.

Table 6: Utility estimates main effects

Attributes Estimates Wald P-value

Product type Grass-fed Grain-fed Soy/lupine-based Vegetable-based 0,7841 0,3853 -0,7247 -0,4447 261,6190 2,0e-56 Production method Organic Non-organic 0,2461 -0,2461 63,9161 1,3e-15 Nutritional claim High protein & low fat High protein Low fat No claim 0,1610 -0,0113 -0,0621 -0,0875 7,5462 0,056 Price €9.50, €10.50, €11.50, €12.50 -0,2123 43,9063 3,4e-11

None option -1,5141 204,65 2,0e-46

4.5.2 Incentive-alignment moderation

(32)

26

Figure 6: Model specifications and corresponding utility estimates

The graph above shows that respondents in the IA condition seem to have a much lower utility for soy/lupine-based hamburgers, although the pattern for the two conditions is the same for this attribute. Furthermore, whereas in the no IA condition respondents appear to be rather indifferent regarding nutritional claims, the utility estimates show more variation in the IA condition. Overall, price seems to have a linear effect on utility in both conditions.

Besides, the model fit of the aggregate model already indicated that price has a linear effect on utility. Therefore, the attribute importance per condition is shown in figure 7 below, including price as a linear attribute.

(33)

27

Figure 7: Attribute importance

Next, in order to formally test the hypotheses, a new model is estimated in which the

interaction terms are included. More specifically, new variables are created that consist of the interaction effect between incentive alignment (0=no IA, 1=IA) and the effect codes of the attribute levels of the type of product, production method, and nutritional claim. As a result, seven new numeric variables are added to the model. First of all, a goodness of fit comparison is made between the main effects (aggregate) model and the model including the interaction effects. As can be seen in table 7 below, the two models have comparable Log-likelihood values. In order to test whether the difference between the two is significant, a likelihood-ratio test is performed. The Chi-Square statistic is 18,6992 with seven degrees of freedom, which results in a critical value of 18,548. Thus, the model with interactions is significantly better than the aggregate model. However, the Log-likelihood always increases with more

parameters. The R2_{-adjusted values and hit rates are the same, which enables continuing with} the interaction model.

Table 7: Goodness of fit comparison

Goodness of fit comparison

Aggregate model: Model with interactions:

LL(0) -1658,008 -1658,008 LL(β) -1288,4220 -1279,0924 Parameters: 9 16 R2 adjusted 0,22 0,22 Hit rate 73,66% 74% Prediction error 26,24% 26%

The parameter estimates and corresponding Wald- and p-valuesare shown in table 8 below. As can be derived from table 8, there is a marginally significant positive interaction effect between IA and grass-fed (utility=0,2384, p=0,057). When the experiment is

incentive-aligned, consumers attach more utility to grass-fed beef compared to the other levels. Another marginally significant positive interaction effect is found between IA and grain-fed

0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% Product type Produc9on

method Nutri9onal claim Price

Softening consumers’ ecological footprint: preferences for meat and meat substitutes

Softening consumers’ ecological footprint:

preferences for meat and meat substitutes

Klasina Holthuis

SOFTENING CONSUMERS’

ECOLOGICAL FOOTPRINT:

PREFERENCES FOR MEAT AND

MEAT SUBSTITUTES

Abstract

Preface

Table of contents

Overview of figures and tables

1 Introduction

2 Theoretical Framework

3 Methodology

4 Results

A1ribute importance

A1ribute importance