The effecTs of socioeconomic sTaTus and healThy-
shopping dynamics on end-of-Trip BaskeT healTh:
a Sequential Market BaSket analySiS
on Scanner Data FroM a Dutch SuperMarket
rutMer FaBer
2
The Effects of Socioeconomic Status and
Healthy-Shopping Dynamics on Basket Health: a Sequential
Market Basket Analysis on Scanner Data From a
Dutch Supermarket
Master Thesis
MSc Marketing Management & Marketing Intelligence
University of Groningen
Author:
Rutmer Faber
Date:
June 20, 2016
Address:
Spinozaweg 68 bis,
3532 SH Utrecht, The Netherlands
Phone number:
+31(0) 644005910
Email address:
r.h.faber@student.rug.nl
Student number:
19079999
3
Table of Content
Management Summary ... 5 Preface ... 6 1. Introduction ... 7 1.1 Research partner ... 8 1.2 Research questions ... 9 1.3 Relevance ... 10 1.4. Outline ... 11 2. Theoretical framework ... 12 2.1 Conceptual model ... 122.2 Existence and definition of Healthy-Shopping Dynamics ... 13
2.3 Psychological theories that contribute to the explanation of healthy shopping dynamics ... 14
2.3.1 Self-regulation theory ... 14
2.3.2 Licensing ... 15
2.3.3 Emotions ... 16
2.3.4 Self-regulation, licensing and emotions in the context of Healthy-Shopping Dynamics 17 2.4 SES affecting Healthy-Shopping Dynamics... 19
3. Methodology ... 25
3.1 Data collection ... 25
3.2 Criteria and sample ... 25
3.3 Operationalization of variables... 26
3.3.1 Healthy shopping dynamics and end-of-trip basket health ... 26
3.3.2 Health Promotion and SES variable ... 28
3.3.3 Purchase decision related variables ... 28
3.4 Research Method ... 29
4. Results ... 34
4.1 Exploratory analysis ... 34
4.2 Effects of SES and HPromDum on mediating variables... 36
4.2.1 Checking model assumptions ... 40
4.3 Direct effects on end-of-trip basket health ... 42
4.3.1 Direct effects of SES and Healthy-Shopping-Dynamics mediators (AveEXtrPD, DWstat) ... 43
4.3.2 Direct effects of the other mediating variables (%HealthyPD, %UnhealthyPD, AvePRicePD, %DiscountedPD and AvePricePD). ... 43
4.3.3 Direct effects of NrFoodPD and HPromDum and interactions with SES ... 44
4.3.4 Checking model assumptions ... 44
4
4.5 Total effects SES and HpromDum on end-of-trip basket health ... 48
5. Conclusion ... 50
5.1 Discussion ... 51
5.1.1 Effects of SES through Healthy-Shopping-Dynamics on end-of-trip basket health ... 51
5.1.2 Effects of SES though healthy and unhealthy purchase decisions on end-of-trip basket health……….. ... 52
5.1.3 Effects of SES through the average food price and discount percentage of purchase decisions end-of-trip basket health ... 53
5.1.4 Effects of the ‘Variatie = Voordeelweken’ on end-of-trip basket health ... 53
5.2 Limitations and further research ... 54
5.3 Managerial implications ... 56
References ... 58
5
Management Summary
The problem of worldwide obesity is already for decades an important point of discussing for many governments and organizations. Overweight amongst consumers is largely driven by the intake of high energy dense and food containing poor nutrients (James, 2008; Asfaw, 2011). This implicates that supermarkets play an important role in improving the health of consumers. Therefore, it is essential to understand how in-store purchase decisions are made and can be influenced. Several researches show differences in health choices due to differences in socioeconomic status (SES), while other researches have shown that in-store decisions are dependent on previous decisions and that health choices evolve during a shopping trip. Research of Waterlander et al., (2012) has shown that Healthy-Shopping Dynamics can negatively influence the effectiveness of health promotions on end-of-trip basket health.
This study tries to demonstrate the presence of healthy shopping dynamics in a real life shopping environment, its interaction with different levels of SES and the effects on the end-of-trip basket health. Dutch supermarket chain Plus Retail has provided scanner data of consumer purchases from several stores that differ in the SES of their primary shoppers. Multiple mediation models are used to assess the effects of SES through indicators of Healthy-Shopping Dynamics and other variables on end-of-trip basket health. Furthermore the effect of health promotions is tested by comparing the effect of one week of the ‘Variatie = Voordeelweken’ against two others weeks outside the promotion period on end-of-trip basket health.
Results of this study show that consumers in stores with primary shoppers with a high SES, on average eat more healthy due to a higher percentage of healthy purchase decisions and a lower percentage of unhealthy purchase decisions. It also appears that they show more variability and extreme shifts between health of sequential purchase decisions. Furthermore, it seems that on average in stores with primary shoppers with a low SES, consumers make increasingly more unhealthy purchase decisions during later parts of the shopping trip. There is no evidence that in stores with a customer base with low SES, consumers follow strategies to purchase cheaper products and therefore make less healthy purchase decisions.
Furthermore, results show that health promotions possibly can have a negative effect on end-of-trip basket health through a decrease in the percentage of healthy purchases and an increase in the percentage of unhealthy purchase decisions. However, this effects can be caused by the operationalization of the end-of-trip basket health variable in this research and divergent purchase behavior in week 1 of the data that is used. There is no statistical evidence that stores who differ in the SES of their primary shoppers moderate the influence of health promotions on end-of-trip basket health.
Keywords: Healthy-Shoppping Dynamics, socioeconomic status, healthy purchase decisions, unhealthy
6
Preface
This research is the last piece of work of my Master Marketing Management & Marketing Intelligence. During the two years that this master educated me in both marketing management and marketing intelligence, I discovered how much I like combining both marketing areas. Over the last years I gained interest to work in the FMCG industry. After an internship at PepsiCo, that was focused on marketing management, I wanted to gain practical experience in the field of marketing intelligence. I am very grateful that this research project came on my path. During the past months I have solved many analytical challenges and I have spent a lot of time on thinking how this research could help Plus in a practical way. The combination of combining both marketing areas and a sneak peak in the retail side of the FMCG industry has learned me many new things and brought me much enjoyment.
I am grateful towards Plus for providing this opportunity and I admire the fact that they really bring their mission of providing good food for their customers into practice. I want to thank Mr. Marco Maatman and Ms. Astrid Westerveld for all the useful discussions and hours that they have put into this research. Furthermore, I want to thank Prof. Dr. ir. K. van Ittersum and Prof. Dr. T.H.A Bijmolt for their valuable feedback and discussions during the process of writing this thesis. Furthermore, I want to thank my research partner Linda Grondsma for her collaboration during the process of setting up the research, data preparation and discussions during the further process. Lastly, I want to thank my parents and my girlfriend Rosanne for their support during the research.
7
1. Introduction
Obesity has a large effect on personal well-being and societal costs, therefore it is an important subject for many governments and organizations (Cawley, & Meyerhoefer, 2012). Obesity results in large scale risks for morbidity and mortality, mainly caused by type two diabetes and atherosclerotic cardiovascular disease, which can prompt strokes and heart attacks (Mokdad, 1999; Willett, Dietz & Colditz, 1999). Lim et al. (2010) showed that estimations for overweight and obesity causing deaths are as large as 3.4 million, whereas obesity also results in a loss of 3.9% of years of life and an increase in disability adjusted life years of 3.8%. Due to the health risks associated with obesity, its rise has been followed with a large scale implementation of obesity prevalence programs all around the world. According to Ng et al. (2013), from 1980 to 2013 the body mass index (BMI) of 25 and greater (overweight indicator) increased from 28.8% to 36.9% for men, and from 29.8% to 38,0% for women. Moreover, prevalence measures of overweight and obesity have increased with 27.5% for adults, and 47.1% for children in this period.
Obesity and overweight amongst consumers is largely driven by consumption of food that is energy dense and contains poor nutrients such as food high in (saturated) fat, sugar and salt (James, 2008; Asfaw 2011). Obesity is also linked to socioeconomic status (SES). Lower levels of education and income have the highest rates of obesity due to inequitable access to healthy foods (Drewnowski, 2003). Appelhans et al. (2012) show that consumers with a low SES purchase cheaper products that contain more calories, fat, and are less nutrient-rich.
According to Regmi & Gehlhar (2005), 52,4% of food expenditures are done in supermarkets on a global level, with the highest percentages observed in Western Europe and The United States (55,9% & 62,1%). Thus, in-store decision making in supermarkets plays an important role in the health of consumers’ shopping baskets, and eventually in the health of society as a whole. To foster healthier food consumption, it is important to understand how food choices in grocery stores can be influenced (Payne, Niculescu, Just & Kelly, 2014). Prior research about in-store decision making has mainly focused on an overall trip-level approach (Gilbride, Inman, & Stilley, 2015). In these researches supermarket purchases have been seen as a sequence of independent purchases, resulting in the fact that we know very little of how healthiness of shopping baskets comes about.
8 number of calories in the whole basket increased, while the share of healthy foods remained constant. This suggests that consumers used the monetary advance from the discounted healthy products to buy unhealthy products. Other researches provide evidence that psychological theories such as ‘licensing’ (Khan & Dhar, 2006), and ‘self-regulation theory’ (Hoyer, Maclnnis & Pieters, 2013) influence later decisions through prior decisions in the shopping process.
Previous information suggests Healthy-Shopping Dynamics influence the healthiness of purchases during a shopping trip. However, little is known about these Healthy-Shopping Dynamics work and they have not yet been researched on real-life sequential purchase data. Analyzing Healthy-Shopping Dynamics and factors that influence them is important in order to be able to understand how health purchase decisions exactly are being made and can be influenced during a shopping trip. Without exact knowledge of these Healthy-Shopping Dynamics and factors influencing them, prevalence measures to improve consumer diets by retailers and policy makers may be inefficient or even backfire, such as research of Waterlander et al. (2012) has shown. Moreover, the influence of other factors such as SES on Healthy-Shopping Dynamics has not yet been researched. This research will focus on how healthiness of shopping baskets comes about and if and how SES influences this process.
Research on Healthy-Shopping Dynamics will be done in this study by analyzing self-scanner data from three different supermarket stores of PLUS Retail in several weeks. Secondary data from PLUS retail on SES of primary shoppers in different stores was used to select three stores that differ significantly in SES of their primary shoppers. This data is used to analyze the scope of this study: the existence of Healthy-Shopping Dynamics and the influence of SES on these dynamics.
1.1 Research partner
This study is done on behalf of Plus Retail, a Dutch supermarket chain that is currently the 5th largest in The Netherlands. Plus’s organization exists mainly from independent supermarket entrepreneurs. Currently there are 218 Plus entrepreneurs, which together own 255 stores (Plus Retail, 2015a). Each store is managed by an independent store owner, who can to some extent adapt the store management to local preferences. The main brand values of Plus are: attention, quality, local & responsibility (Plus Retail, 2015b). In 2015 and 2016 PLUS was ranked as best Dutch corporate social responsible supermarket in a report by GfK (Plus Retail, 2015c). Moreover, PLUS received excellent scores on providing information about food nutrition for their customers.
9 healthy food for children. In 2015 Plus started with the ‘Variatie = Voordeelweken’ campaign, this promotion lasts for three weeks and offers a monetary discount if a certain variance of healthy products is purchased. Moreover, the importance of eating a variance of healthy food products is emphasized during the campaign period.
1.2 Research questions
Most existing studies on health choices are focused on improving the healthiness of single product purchases (Chandon & Wansink 2012; Payne et al. 2014). Different factors in these researches such as price (e.g. pricing strategies), marketing communication (e.g. branding and food claims), packaging (e.g. design and package based claims), environment (e.g. availability, salience and convenience) are proven to influence our purchases and food intake (Payne et al. 2014). Waterlander et al. (2012) show that strategies stimulating a healthy diet may be eradicated when analyzed as part of a larger shopping trip. Therefore, it is essential to demonstrate presence of Healthy-Shopping Dynamics, and gain understanding of evolvement of these dynamics during the shopping trip.
There is also evidence that SES effects Healthy-Shopping Dynamics. Research of Appelhans et al. (2012) has shown that there is a difference in health of products that high and low SES consumers purchase. Wardle & Steptoe (2003) associate lower SES with less health consciousness, stronger believe of chance affecting health, less future oriented- and lower life expectancies (e.g. less health motivated). The difference in health motivation between different levels of SES in the research suggests an effect of SES on self-efficacy and self-regulation regarding the health of purchasing decisions. Consumer with a different SES might experience differences in self-regulation (Winett, & Wojcik, 2007), which in terms may lead to differences in Healthy-Shopping Dynamics. Furthermore, several researches show that consumer with a low SES have a more tight budget for food expenditures (Drewnowski, & Specter, 2004; Pechey & Monsivais; 2015). It could be that consumers with a lower SES buy cheaper food, which in terms influences the end-of-trip basket health.
The first purpose of this study is to address the existence of Healthy-Shopping Dynamics in real-life sequential purchase data, and to address the effect on the healthiness of the end-of-trip shopping basket. The second purpose of this study is to research the influence of SES on Healthy-Shopping Dynamics and end-of-trip basket health. The corresponding research question to address both issues is:
What are the effects of Healthy-Shopping Dynamics, SES, and the interaction of
both on the health of the end-of-trip shopping basket?
10 explained in the context of Healthy-Shopping Dynamics. Moreover, it is important to define what Healthy-Shopping Dynamics are and how they can be measured. It is also important to gain understanding of theoretical concepts through which SES can influence Healthy-Shopping Dynamics and end-of-trip basket health. Therefore, the following sub-questions to address these issues will be answered in the theoretical framework.
1) How can Healthy-Shopping Dynamics be defined and measured?
2) Which theoretical concepts can attribute to the understanding of Healthy-Shopping Dynamics?
3) Which theoretical concepts can explain the effect of SES on Healthy-Shopping Dynamics? 4) How can health promotions decrease the total health of the end-of-trip shopping basket? 5) How does SES interact with drivers of Healthy-Shopping-Dynamics and healthy and unhealthy
purchases and what effects has this on end-of-trip basket health?
6) How does SES interact with strategies to purchase cheaper products and what effect has this end-of-trip basket health?
The existence of Healthy-Shopping Dynamics and the effect of SES of Healthy-Shopping Dynamics can be verified when all variables and concepts that play a role in the interplay between SES and Healthy-Shopping Dynamics are identified, and the total model and the impact of its variables is tested. The insights gained from this model will contribute to answer the research questions in this study.
1.3 Relevance
This study aims to establish the existence of real-life Healthy-Shopping Dynamics and the effects of SES on these dynamics and end-of-trip basket health through the analysis of sequential purchase data from supermarket self-scanners. The use of scanner data has multiple advantages over conventional data as the data is very accurate, related directly to the sales act of an individual customer, and provides information of the competitive environment of the customer decisions (Guadagni & Little, 2008).
11 Besides contributions to the academic research field, the study will provide useful insights for PLUS Retail regarding their how to stimulate healthy-purchase decisions amongst consumers. With this research, PLUS wants to generate consumer insights in Healthy-Shopping Dynamics and eventually find manners to improve the overall health of their consumers shopping basket. Plus stores are located throughout the country and have different customer demographic profiles. Some stores are located in areas with predominantly low SES consumers, whereas other stores are located in areas with predominantly average or high SES consumers. Differences between these consumers groups can provide manners for different approaches to improving the healthiness of consumers shopping basket in these particular areas.
1.4. Outline
12
2. Theoretical framework
This chapter starts with a visual representation of the conceptual model. In the following section the definition of Healthy-Shopping Dynamics is established. Next, relevant academic work that provides insights in psychological theories and contributions to the understanding of the phenomenon is presented and explained. Building on these theories and contributions, the last part of the theoretical framework will explain effects of SES on Healthy-Shopping Dynamics and end-of-trip basket health and generate hypothesis.
2.1 Conceptual model
* Healthy-Shopping Dynamics are defined as shifts in consumer decisions regarding the health of products being purchased during the shopping trip
Extremeness of Healthy-Shopping Dynamics *
13
2.2 Existence and definition of Healthy-Shopping Dynamics
Several researches have shown that prior decisions and choices in the shopping process influence subsequent decisions in the shopping process (Khan & Dhar, 2006; Vohs & Faber, 2007; Waterlander et al, 2012; Gilbride, Inman & Stilley, 2015). Dhar, Huber & Kahn (2007) show that there exists a shopping momentum in a consumers brain, wherein after an initial spending decisions a mental barrier is crossed, which leads to the purchase of additional items. Mechanism behind this theory is, that a first purchase shift the consumers mindset from deliberate-browsing to implementation based shopping (Dhar, Huber & Kahn, 2007). Vohs & Faber (2007) find that after an initial purchase, consumers are less able to control future spending. The researchers attribute this finding to depletion of the self-regulatory resources in a consumer’s brain, which increases the probability of impulse (unplanned) buying behavior. Gilbride et al. (2015) find that the likelihood of unplanned purchases can evolve in a dynamic way and differs across consumers with different trip budgets. Moreover, research of Waterlander et al. (2013) has shown that similar dynamic effects may be present in healthiness of purchase decisions. Van Ittersum & Bijmolt (2015) conducted a pilot study to gain insights in Healthy-Shopping Dynamics. In this pilot study 54 MTurk participants were asked to pretend they were grocery shopping and were about to make 11 purchases. Each purchase decision had four options and a no-purchase option, options were accompanied by a picture of the product, its price, and caloric information (Appendix A1). Healthiness in the study was operationalized by the relative number of calories selected from the choice set. Results of the pilot study did reveal preliminary evidence for Healthy-Shopping Dynamics through a non-linear pattern in the health of purchase decisions. Results suggest that shoppers make relative healthy choices early in the shopping trip, followed by relative unhealthy choices during the second half of the shopping trip (Appendix A2). This is consistent with findings of Khan & Dhar (2006), that show an increased probability for self-indulgent options after prior purchases that boost self-concept. Additional analysis of the pilot study has shown that the pattern that is found, is especially present amongst overweight shoppers.
14
2.3 Psychological theories that contribute to the explanation of healthy shopping dynamics
There are several psychological theories that can contribute to the explanation of Healthy-Shopping Dynamics. The aim of this research is not to define exactly how these different theories work together in the light of Healthy-Shopping Dynamics. Therefore, the factors in the following paragraph are not measured or processed in the mathematical model. However, is it important to understand these theories, to better understand Healthy-Shopping Dynamics and to form hypotheses on how SES influences Healthy-Shopping Dynamics. The first sections of this chapter will explain the psychological concepts that possibly play a role in the existence of Healthy-Shopping Dynamics. These findings are later used in the sections that describe possible differences in the influence of SES on Healthy-Shopping Dynamics.
2.3.1 Self-regulation theory
For Healthy-Shopping Dynamics to exist, there must be a significant number of consumers that are in some way health motivated. Such consumers can be seen as consumers exerting a goal wherefrom the desired outcome is remaining or becoming healthy. Reaching such goal requires self-control, which can be defined as the exertion of control over yourself. A person exerting self-control attempts to change the way he or she would otherwise think, feel, or behave (Muraven & Baumeister, 2000). According to Hoyer, Maclnnis & Pieters (2013, p. 56) consumers seeking to exert self-control are caught in a psychological conflict between desire, which is a short-term, hedonic force (we want that candy now, even if we feel regret later), and willpower, which is a long-term, more utilitarian force (we think and act to stop ourselves from having that candy now, to have a long-term healthy lifestyle). The former leads to conclude that self-regulation or self-control plays an important role the decision to purchase healthy or (more) unhealthy products for health motivated consumers.
According to Baumeister & Heatheron (1996) self-regulation is built on three components. Initially, a goal or a set of standards must exist to reach a desired state. Such as the goal or desire to remain or become healthy. Second, the current state and the desired state must be monitored till some extent. Monitoring in the context of Healthy-Shopping Dynamics exists of the comparison of the planned healthiness of the shopping basket to the current healthiness of the basket. The third component refers to the behavior change, when the current state is evaluated negative compared to the desired state (Baumeister & Heatheron,1996).
15 a high amount of glucose, when adding glucose during depletion of the self-control resource, the potential for self-control increases. Tice, Baumeister, Shmueli & Muraven (2007) found that the induction of a positive mood could reverse depletion. Moreover, Schmeichel and Vohs (2009) found that self-affirmation had a similar effect and could replenish self-regulatory recourses.
Other researches show that previous explanations of the self-regulation theory may not cover the full working of depletion effects. There is proof that similar decision situations (e.g. deciding between healthy and unhealthy food) may actually increase self-control resources (Dewitte, Bruyneel & Geyskens, 2009). Converse & DeShon (2009) attribute these findings to adaption and learning processes. Their research shows that in situations wherein opportunities to adapt are sufficient (e.g. completing two initial tasks), adaption may occur, resulting in increased self-control resources. In situations or research where there is not enough opportunity to adapt (e.g. completing one initial task, such as in many academic research on self-control), no chance is provided for adaptation processes to develop and instead depletion effects are observed. However, It is unclear, how these two processes unfold over substantially longer periods of time. For example, “the depletion effect might continue to grow, whereas the adaptation effect might remain stable, resulting in a return to depletion results” (Converse & Deshon, 2009, p. 1323). Researches that contribute to support this theory find increased probability for unplanned or unhealthy food later on in a shopping trip (Stilley, Inman & Wakefield, 2010a; Gilbride et al., 2013; Gilbride et al., 2015)
The elements of the self-response model suggest that when a consumer makes a healthy purchase (e.g. choosing between less or more healthy options), this will increase the probability of a subsequent unhealthy purchase due to depletion of the self-control strength resources. However, similar choices such as controlling for healthy purchases may actually increase self-control resources (Dewitte et al., 2009; Converse & Deshon, 2009). Therefore, it is likely that self-regulation on healthy-shopping decisions may need several decisions before adaption effects are taken over by depletion of the self-regulatory resource, and consumers give in to their need for a more indulgent choice.
2.3.2 Licensing
16 De Witt Huberts, Evers & de Ridder (2012) shows that self-licensing can result in an increased consumption of hedonic goods (e.g. unhealthy food), while ruling out other factors such as: resource depletion, negative affect, and visceral state as alternative explanations. Thus, licensing can be seen a separate mechanism, that can lead to increased hedonic consumption (e.g. unhealthy food), independent of other factors.
Licensing effects are in conflict with a person’s motivation for consistency, which can be seen as a persons need to be consistent in knowledge, opinions, or beliefs about the environment, about oneself, or about one’s behavior (Gawronksi, 2012). According to Mullen & Molin (2016) individuals are more likely to exhibit consistency when they focus abstractly on the connection between their initial behavior and their values, whereas they are more likely to exhibit licensing when they think concretely about what they have accomplished with their initial behavior.
2.3.3 Emotions
According to Wilcox, Kramer, & Sen (2011) Emotions play an important role in a consumer’s goal pursuit. Emotions can be classified into two categories: basic emotions and self-conscious emotions. According to Tracy & Robins (2004), basic emotions are biological based and can be studied without verbal information (e.g. anger, fear, disgust, sadness, happiness, and surprise). Self-conscious emotions (e.g. pride, guilt, shame, and embarrassment) reflect people’s response to own behavior or characteristics, and require a certain level of self-representation, and generate self-evaluation (Tracy & Robins, 2004). According to Tangney & Fischer (1995), self-conscious emotions influence motivation and regulation of a person’s thoughts, feelings, and behaviors. Moreover, Giner-Sorolla (2001) shows that self-conscious emotions can support self-control if they or brought into mind before the chance to act.
Mukhopadhyay & Johar (2009) show that a decision to buy or not to buy evokes different affective responses. Resisting an unhealthy purchase, by instead purchasing a healthy product, may evoke feelings of incidental pride. According Tracy, Shariff, & Cheng (2010), pride is a conscious emotion, which is involved in both self-awareness and self-assessment. Wilcox et al. (2011) show that pride can influence long-term goal pursuit through conflicting processes. When consumers feel pride in a sense of achievement, they show an increased probability to make indulgent choices (licensing). However, when they feel increased self-awareness, they show an increased probability to make more virtuous choices. Emotions of pride have proven to be effective in facilitating self-regulation and the ability to pursue certain goals. Williams & DeSteno (2008), show in their experiment that pride facilitates self-regulation and the ability to pursue goals, despite short-term losses.
17 individual must attribute the cause of the event to the self, resulting in blaming his or her self for the situation. Individuals will experience shame if they blame poor performances on ability, which is an internal stable factor, whereas individuals will experience guilt if they blame their poor performance on effort, which is an internal unstable factor (Tracy & Robins, 2004). In the process of grocery shopping it is less likely to experience shame after choosing a self-indulgent option, because for almost every product there is an ability to choose for a more healthy option. However, if a consumer is unable to resist a temptation, and purchase a relatively unhealthy product they may experience guilt.
Giner-Sorolla (2001) shows that guilt from indulging into unhealthy food stimulates people to secure their self-esteem and alleviating guilt by reducing subsequent indulgence. According to Chen & Sengupta (2014) choosing a vice (e.g. tasty but unhealthy product) can result in guilt and accordingly in lower vitality (e.g. psychological energy). Lower vitality in turn will have a negative impact on self-control (Muraven, Gagné, & Rosman, 2008). In contrast, the consumption and purchase of unhealthy purchases (e.g. vices) can also increase vitality (e.g. psychological energy) due to the intrinsic pleasure afforded by the unhealthy product (Chen & Sengupta, 2014). This increased vitality will result in increased self-control (Muraven et al., 2008). The latter effect in the research of Chen & Sengupta (2014) was only found to appear in situations wherein consumers experienced low autonomy. Assuming that consumers experience high autonomy in most cases during grocery shopping, it is more likely that experiencing guilt results in lower self-control.
The decision to choose a vice or multiple vices can also result in goal abandonment through an emotional response called the ‘What-The-Hell’ (WTH) response (Cochran & Tesser, 1996, p. 99). This effect refers to a motivational shift, where as a result of failure of a sub goal the individual abandons the discipline associated with attempting to meet the long-term goal (Cochran & Tesser, 1996, p. 110). Such long-term goal in the context of Healthy-Shopping Dynamics would be remaining or becoming healthy. Abandonment of such goal may lead to a break-down of the self-regulatory model, wherein an unhealthy purchase would result in a consumer dropping its health goal.
2.3.4 Self-regulation, licensing and emotions in the context of Healthy-Shopping Dynamics
To understand the workings of licensing and self-regulation in the context of Healthy-Shopping Dynamics it is important to understand the differences between both theories. Licensing in the context of healthy-shopping decisions is similar in a way to self-regulation theory, as both theories assume that making a healthy purchase decision or a series of healthy purchase decisions, will lead to an increased probability of choosing a subsequent more indulgent choice. However, Licensing and self-regulation theory are fundamentally different in terms of motivation. A breakdown of self-regulation can be seen as giving in to impulse behavior.
18 of restraint or control of that impulse (e.g. depletion of self-regulatory resources). Impulsiveness has been classified as a personality trait (Goldberg et al., 2006). Licensing differs from self-regulation because people justify their indulgence with emotions or reasoning (Kivetz & Zheng, 2006; De Witt Huberts et al., 2012). In summary, self-regulation theory assumes indulgence due to decreased willpower (ego-depletion), whereas licensing occurs when people feel they deserve indulgence. Thus, licensing effects will only happen when people choose hedonic goods (e.g. vices), and not when choosing utilitarian goods (e.g. healthy products in the context of Healthy-Shopping Dynamics), because the context of the decision results in justifying the decision (Khan & Dhar, 2006; Kivetz & Zheng, 2006). A consumer with a health goal will for example justify the decision to buy chocolate, after several prior heathy purchases.
Emotions have shown to influence both licensing and self-regulation theory (Wilcox et al., 2011; Chen & Sengupta, 2014). Pride can result in both licensing, which results in an increased probability of a subsequent more indulgent purchase, and increased self-regulation, which results in an increased ability to pursue ones health goal. Guilt in the light of Healthy-Shopping Dynamics is most likely to have a negative effect on self-regulation, which leads to a higher probability to give in to indulgence (Chen & Sengupta, 2014). However, when guilt is too high, it is likely that a health motivated consumer wants to secure their self-esteem and alleviate this guilt (Giner-Sorolla, 2001). This can be done shifting purchase decisions in favor of more healthy options. Lastly, the health motivation goal can also be thrown overboard in a WTF response (Cochran & Tesser, 1996). This response results in a breakdown of the self-regulatory model, and an increased probability for the purchase of more indulgent options.
19
Health motivated consumer make a(n)…
Followed by a(n)… Healthy choice Unhealthy choice
Healthy choice Pride, Increased self-regulation (through pride or adaption effects)
Guilt, increased self-regulation (through prior unhealthy choice)
Unhealthy choice Licensing, decreased self-regulation (through prior healthy choice),
Decreased self-regulation (through guilt), WTH response
Table 1: Potential drivers of Heathy-Shopping Dynamics among health-motivated consumers 2.4 SES affecting Healthy-Shopping Dynamics
According to Galobardes, Shaw, Lawlor, Lynch & Smith (2006), SES is based on the position of a person in society according to social and economic factors such as income and education. There are several researches that relate differences in SES to consumer health. Wardle & Steptoe (2003) show that lower SES is related lower health motivation. Lower SES leads to less health consciousness, more belief of luck affecting health, less future orientation and lower life expectancies. This implies that self-efficacy of consumers with a low SES is lower than self-self-efficacy of consumers with a high SES. Individuals with higher self-efficacy and more belief in favorable outcome expectations have a higher probability of eating healthier trough implementing self-regulatory behaviors (Bandura, 2004). Appelhans et al. (2012) show in their research that that consumers groups differing in SES show differences in costs and ingredients of purchased products. Results of the research indicated that consumers with a lower SES spend less money on 100 kcal food that consumers with a higher SES, while lower energy costs resulted in more fat, and less proteins, dietary fiber and vegetables per 1000 kcal.
There is also research that suggest that the food environments of different neighborhoods may drive socioeconomic differences in diet quality. The reasoning for this is that more deprived consumers often live in areas that have more unhealthy food outlets than healthy outlets, and thus have less physical access to healthier food (Cummins et al., 2009; Moladi, Leyland, Ellaway, Kearn & Harding, 2012). However recent research of Pechey & Monsivais (2015) shows that supermarket choice and shopping behaviors do not contribute so socioeconomic differences, and thus, that supermarket environments are not likely to influence the relationship between supermarket choice and health of purchases. Following the reasoning that SES is positively related to healthiness of purchased products, without influence of physical access to healthy food, it is likely to assume that SES is positively related to end-of-trip healthiness of the shopping basket.
H1: SES has a positive effect on end-of-trip basket health.
20 grains, fat and added sugar have a high energy density, and are cheaper than healthier food comprised of fish, lean meats, fresh vegetables and fruits. Clinical studies also show that high energy density and palatability of sweets and fats are associated with higher energy intake (Drewnowski, & Specter, 2004). Drewnoski (2009) concludes that when income drops, energy dense and poor nutrient food, is the best way to provide daily calories at an affordable cost.
In research of Anderson et al., (2007) self-regulation was found the strongest predictor of healthier nutrition. Furthermore, they also found a direct positive effect between SES and self-regulation. Thus, lower SES is associated with lower self-regulation, which can lead to an accelerated increasing probability of indulgent options. Moreover, Anderson et al. (2007) found an effect of SES through the intake of fruits and vegetables. A higher SES leads to the intake of more fruits and vegetables, which was also associated with higher self-regulation. Self-regulation can be reflected by withholding the purchase of indulgent options, it can thus be expected that higher levels of SES are purchasing less indulgent options compared to lower levels of SES, which influences the end-of-trip basket health. Thus, it is likely that higher levels of SES make more intrinsically healthy and less intrinsically unhealthy purchase decisions.
H2: Healthy purchase decisions have a positive effect of on end-of-trip basket health.
H2a. The positive effect of SES on the end-of-trip basket health is mediated by the percentage of healthy purchase decisions.
H3: Unhealthy purchase decisions have a negative effect on end-of-trip-basket health.
H3a: The positive effect of SES on the end-of-trip basket health is mediated by the percentage of unhealthy purchase decisions.
Several researches how that consumer with a low SES have a more tight budget for food expenditures (Drewnowski, & Specter, 2004; Pechey & Monsivais; 2015). According to Jetter & Cassady (2006) as expenditures for a healthy diet can constitute up to 40% of consumers with a low income. As low cost foods, with a high energy density, are cheaper than healthier foods, it can be expected that consumers with a low income provide daily calories at an affordable cost (Drewnowski, 2004). The average price of food products is expected to have a positive effect on end-of-basket health. Consumers with a low SES compared to a higher SES consumers are expected to buy on average cheaper products, who’s effect is more negatively related to end-of-trip basket health.
21 H4a: The positive effect of SES on the end-of-trip basket health is mediated by the average price of food purchase decisions.
Another strategy to achieve a similar outcome is being more discount oriented. Consumer with a low SES have to pay more attention to food prices. Therefore, it might be likely that they are more discount oriented. It is not likely that the discounted products would be on average more healthy of unhealthy than products who are not in discount. However, it might be that Plus balances their discounts in favor of more healthy products due to their focus on ‘good food’ for their customers. As it is expected that SES differs in the percentage of products bought in discount, it is worth checking this influence on end-of-trip basket health.
H5: The percentage of discounted purchase decisions has an effect on end-of-trip basket health.
H5a: The positive effect of SES on the end-of-trip basket health is mediated by the percentage of discounted purchase decisions.
SES is negatively related to obesity (Drewnowski, 2003; Mclaren, 2007; Drewnoski 2009; Appelhans et al. 2012). Herman & Polivy (2008) show that external food cues, and especially palatable food cues have a specific strong effects on consumers with overweight (e.g. obese consumers). Resulting in the fact that overweight consumers are more tempted to indulge during the shopping process. Effron, Monin, & Miller (2013) show that tempted consumers (e.g. dieters) are likely to exaggerate the sinfulness of previous not taken decisions, and therefore creating an increased illusion that they refrained from indulgence. This eventually results in an increased probability to license future indulgence.
22 Concluding, it is possible that lower SES leads to an accelerated shift between choosing for unhealthy products after healthy products and vice versa.
H6: SES has a negative effect on the number of shifts between the purchase of healthy and unhealthy products and vice versa (e.g. number of waves / shifts or variability).
Reasoning above can also be used in predicting the extremeness of shift in healthiness between the decision shift for healthy to unhealthy products and vice versa. As low SES is related to obesity, these consumers are specifically focused on eating healthy, because of their increased of consumers with overweight. On the other hand they are confronted with a strong urge to fulfill the goal of getting enjoyment from eating. As the most unhealthy products often lead to most eating enjoyment, it is likely to believe that consumers with low SES show a more extreme pattern in healthiness of purchases. Thus, an increased extremeness in healthiness between the shift from purchasing healthy to unhealthy products and vice versa.
H7: SES has a negative effect on the extremeness (variability) of the shifts from between healthy and unhealthy products and vice versa (e.g. more extreme amplitude in the waves). As there is a direct positive effect on end-of-trip basket health (Drenowski & Specter, 2004; Drenowski, 2004; Appelhands et al., 2012), and a likely difference between SES on the number and extremeness of shifts as drivers in Healthy-Shopping Dynamics, it might also be that the effect of SES on end-of-trip basket health is mediated by Healthy-Shopping Dynamics.
H8: The positive effec of SES on end-of-trip basket health is mediated by the number of shifts and the extremeness of shifts between the purchase of healthy and unhealthy products and vice versa.
Research of Waterlander et al. (2013) has shown that the promotion of healthy products may have a negative influence on the end-of-trip healthiness of shopping baskets. Effron et al., (2013) show that especially tempted consumers, who are likely to be overly present in consumers groups with a low SES, are likely to engage in indulgence because they exaggerate the sinfulness of previous not taken actions. It seems likely that such consumers would be also tempted to exaggerate the healthiness or positive effect of previous healthy products decisions to indulge. This leads to believe consumers with lower SES are more likely to license indulgence choices after prior healthy decisions than consumers with a higher SES. Thus, healthy product promotions might have a larger negative effect, for low SES consumers compared to high SES consumers, on end-of-trip basket health.
23 H9a: The negative effect of healthy product promotions on end-of-trip basket health is negatively moderated by SES.
As healthy product promotions, as the result of Healthy-Shopping-Dynamics, have a negative effect on end-of-trip basket health, it is likely that an increase in healthy purchase decisions is outweighed by an increase in subsequent unhealthy purchase decisions due to licensing effects. Therefore, it can be expected that the promotion of healthy products increases the percentage of healthy purchase decisions and the percentage of unhealthy purchase decisions. It is expected that the latter effect is stronger than the effect of the increase in the percentage of healthy purchase decisions, which results in a negative effect on end-of-trip basket health.
H10: The negative effect of health promotions on end-of-trip basket health is mediated by the percentage of healthy purchase decisions.
H11: The negative effect of health promotions on end-of-trip basket health is mediated by the percentage of unhealthy purchase decisions.
Lastly, it is important to control for the number of food purchase decisions, as this variable influences the effects of several variables in this research. It is expected that the number of food related purchase decisions is negatively related to end-of-trip basket health.
24
Hypothesis
H1: SES has a positive effect on end-of-trip basket health
H2: Healthy purchase decisions have a positive effect of on end-of-trip basket health.
H2a. The positive effect of SES on the end-of-trip basket health is mediated by the percentage of healthy purchase decisions.
H3: Unhealthy purchase decisions have a negative effect on end-of-trip-basket health.
H3a: The positive effect of SES on the end-of-trip basket health is mediated by the percentage of unhealthy purchase decisions.
H4: The average price of food purchase decisions has a positive effect on end-of-trip basket health. H4a: The positive effect of SES on the end-of-trip basket health is mediated by the average price of food purchase decisions.
H5: The percentage of discounted purchase decisions has an effect on end-of-trip basket health. H5a: The positive effect of SES on the end-of-trip basket health is mediated by the percentage of discounted purchase decisions.
H6: SES has a negative effect on the number of shifts between the purchase of healthy and unhealthy products and vice versa
H7: SES has a negative effect on the extremeness of the shift from healthy to unhealthy products and vice versa.
H8: The effect of SES on end-of-trip basket health is mediated by the number of shifts and the extremeness of shifts between the purchase of healthy and unhealthy products and vice versa.. H9: Healthy product promotions have a negative effect on end-of-trip basket health.
H9a: The negative effect of healthy product promotions on end-of-trip basket health is negatively moderated by SES.
H10: The negative effect of health promotions on end-of-trip basket health is mediated by the percentage of healthy purchase decisions.
H11: The negative effect of health promotions on end-of-trip basket health is mediated by the percentage of unhealthy purchase decisions
H12: The number of food purchase decisions has a negative effect on end/of/trip basket health.
25
3. Methodology
This chapter starts with explaining how the data sample is chosen and collected. Hereafter, the operationalization of important variables in the research is explained, followed by model specifications and the research method.
3.1 Data collection
The research is conducted with data of Dutch grocery retailer PLUS. The data in this research comes from self-scanners in three selected Plus supermarkets and is from the beginning of 2016. It contains the exact sequence of purchases made by each individual customer. Furthermore, it contains information about the shopping start and end time, unique product code, and the amount of discount on a purchase. To assess the healthiness of the purchased products, the sequential purchase data is combined with data about the nutrition of the product assortment.
To research the influence of the promotion of healthy products on end-of-trip basket health and the influence of SES on this relation, data from weeks with differences regarding the promotion of healthy products is used. Data from the ‘Variatie = Voordeelweken’ week is compared against data from other weeks with less healthy products in promotion to assess the influence of healthy promotions on end-of-trip basket healthiness and the moderation effect of SES on this relation.
The influence of SES in this research is assessed by looking at the SES of the customer base of three Plus stores. Information about the presence of several personas in each store is used to select three stores with different primary shoppers. The first selected store has an overrepresentation of customers with a high education and an income above average, the second store has an overrepresentation of customers with low education and an below average income, the last store has no overrepresentation of any specific customer segment. The customers from the stores are classified as high, medium and low SES.
The final dataset constitutes of data of three different Plus stores from three different weeks, with differences in the promotion of healthy products and SES of the primary shoppers.
3.2 Criteria and sample
26 According to Cohen (1992), a statistical power of .80 is conventional to prove a certain relationship between variables. This means that the probability that an existing relationship between variables is detected is 80%. According to Cohen (1992), the expected effect size (ES) of the population must be set before the research. The ES for least-squares regression is defined as F2=R2/1-R2. A medium effect size is specified as an effect likely to be visible by the naked eye of a careful observer, whereas a small effect size is noticeably smaller, but not so small as to be trivial (Cohen, 1992). It is expected that effect sizes in this research are ranging from medium (.15) to small (.02). This leads to an R2 of 0.130 for a medium effects size, and an R2 of 0,196 for a small effect size, when solving the ES equation. The minimum sample size to detect a medium relation for a certain number of predictors is given in Appendix B1. At 8 predictors a minimum sample of 107 respondents is needed to detect a medium effect (Appendix B1). From 6 to 7 and 7 to 8 predictors, the sample size increases with 5. Extrapolating this to 12 predictors, leads to a needed sample of 127 respondents. This is way below the sample sizes used in this study. For a small effect, At 8 predictors a minimum sample of 757 respondents is needed (Appendix B1). From 6 to 7 and 7 to 8 predictors, the sample size increases with 40. Even when the effect is small, the sample is believed to be large enough to reach a statistical power of .80.
A similar minimum sample estimation for a statistical power of .80 for detecting mediation effects is produced by Fritz & Mackinnon (2007). The minimum sample is estimated using the small, medium, and large effect sizes of Cohen (1992), for the paths a1 and b1 in figure 1. It is expected that t the effect sizes of the paths are small (0,02) According to Fritz & Mackinnon (2007),If both paths reach at least a medium effects size, the minimum sample must be at least 78, if both paths reach a small effect size, the minimum sample must at least be 558 (Appendix B2). Even when one of the paths is smaller than expected, It is not expected that the minimum number of respondents needed lies above 900 respondents to reach statistical power of .80, as the minimum for a small and large effects size path is 558, and the difference from a medium to a small effect size in path a1 requires ‘only’ 154 more respondents (Appendix B2).
3.3 Operationalization of variables
The variables described in this research can be operationalized in different ways. The operationalization of variables is important for the understanding of the outcomes in this research. Therefore, the operationalization of all variables used in this research is described in detail in the following sections.
3.3.1 Healthy shopping dynamics and end-of-trip basket health
27 lean meats, fresh vegetables and fruits (Drewnowsky & Specter, 2004). Therefore, healthiness of purchase decisions is measured in terms of energy density, specifically, in calories per 100 gram of the purchased product. From a psychological point of view a purchase that entails multiple the same products, can be seen as one purchase decision. The data used in this research does not allow to check the different total weights or package sizes of the purchased products. Therefore it is decided to be consistent in weighting the health of each purchase decisions, and judge the healthiness of a purchase decision on the amount of calories per 100 g. Thus, differences in package size and quantities of products purchased are not weighted in the health of a purchase decision. The end-of-trip basket health is calculated by the sum of the calories per 100g of all purchase decisions.
The chosen definition of Healthy-Shopping-Dynamics allows for differences in the number of shifts (e.g. number of waves or variability) and the amplitude (extremeness) of the waves in the pilot study of Van Ittersum & Bijmolt (2015) (Appendix A2). Operationalization of Healthy-shopping dynamics in this research focusses on these two aspects. Differences in the number of shifts between healthy and unhealthy products and vice versa can be captured by the Durbin-Watson statistic, that measures the variability between purchase decisions. After separating the intercept linear trend of the healthiness of the purchases, the error terms remain, which can be used to check autocorrelation (Shehu, Bijmolt & Clement, 2016). Positive autocorrelation Positive autocorrelation means that the residual in t tends to have the same sign as the residual in t−1 (Leeflang, Wieringa, Bijmolt & Pauwels, 2014, p. 130). This indicates that the healthiness of the purchased products moves in the same direction (low variability e.g. les waves), whereas negative autocorrelation means that healthiness of the purchased products moves into opposite directions (high variability e.g. more waves). The Durbin-Watson statistic ranges from 0 to 4. A value close to zero indicates strong positive autocorrelation, a value close to 2 indicates no auto correlation, and a value close to 4 indicates strong negative autocorrelation (Leeflang et al., 2014, p. 131).
Differences in the extremeness of the shifts between healthy and unhealthy products and vice versa can be measured by the average shift in calories per 100 g of the different purchase decisions. The absolute difference between the healthiness in calories per 100g of subsequent purchase decisions is summed and divided by the total number of purchase decisions -1, since the first purchase decision has no reference point. Together both variables cover differences in the number of shifts (e.g. number of waves) and the amplitude (extremeness) of the waves in the outcomes of the pilot study of Van Ittersum & Bijmolt (2015), which are defined in this research as Healthy-Shopping-Dynamics.
Average extremeness per purchase decisions =
28
3.3.2 Health Promotion and SES variable
Data from the ‘Variatie = Voordeelweken’ week is compared against data from other weeks with less healthy products in promotion and no extra marketing communications focused on the importance of healthy eating, to assess the influence of healthy promotions on end-of-trip basket healthiness and the moderating effect of SES on this relation. In the data there is one week in the ‘Variatie = Voordeelweken’ that has more healthy products in promotion the other weeks. The variable that focusses on the promotion of healthy products is a dummy variable with a 1 for the week that shows significantly more healthy product promotions, and a 0 for the two weeks that show less healthy products in promotion.
As previously said, SES in this research is not measured on the individual customer level. There is no information available about the demographics of each individual customer that can be paired with their sequential shopping behavior. However, Plus has conducted research wherein they clustered the most common customers shopping in their supermarkets. One of the personas distinguished in this research has clearly a high SES, this person is higher educated, has an above average salary and is environmentally and health conscious. Another persona has a lower SES, is lower educated has a below average salary is more focused on price and promotions. The customers in the store classified as low SES have an overrepresentation of the latter persona, the customers in the store classified as high SES have an overrepresentation of the former person and the store classified as medium SES has no overrepresentation of any of the different personas distinguished by PLUS. Due to the categorical nature, the variable is included as two dummy variables in the specified model.
3.3.3 Purchase decision related variables
Several variables in the research are related to the purchase behavior of customers and specifically to the different groups of SES specified in this research. Anderson et al., (2007) have shown that there is a difference between different levels of SES and the purchase of healthy and unhealthy products. The variables in the research that check this assumption are related to the purchase of products in intrinsically healthy of unhealthy product categories. For healthy purchases this are purchase decisions in the category AGF, that includes fruits, vegetables and potatoes. For unhealthy purchases this are purchase decisions in the category of chocolate, cookies or candy products. Dummies have been made for purchase decisions in either intrinsically healthy or unhealthy product categories. The sum of these dummies has been used to calculate the total number of unhealthy and healthy purchase decisions for each customer. Subsequently the variables are transformed into the percentage of unhealthy and healthy purchase decisions of the total purchase decisions for each customer.
29 Percentage of unhealthy purchase decisions =
Total number of purchase decisions in candy, cookies or chocolate Total number of purchase decisions
As healthy products are often more expensive it is expected that the average price of food products has a positive influence on end-of-trip basket health. The average price of food is also calculated per purchase decision, in which quantity and weight or size of the product or not weighted. The average price per food purchase decision for each consumers is calculated by the sum of the prices of all food purchase decisions and subsequently divided by the total number of purchase decisions.
Average price per purchase decisions = Total price of all purchase decisions Total number of purchase decisions
The percentage of discounted purchase decisions per consumer is operationalized in a similar way as the previous variables. It is the percentage of purchase decisions that included a discount. First, dummies are made for purchase decisions involving a discount. The sum of these dummies has been used to calculate the total number of discounted purchase decisions. Subsequently the variable is transformed into the percentage discounted purchase decisions of the total purchase decisions.
Percentage of discounted purchase decisions = Total number of discounted purchase decisions Total number of purchase decisions
3.4 Research Method
The objective of this study is to examine the presence of healthy shopping dynamics, its interaction with different levels of SES, and the effects on the end-of-trip basket health. The model that includes this effects consists of multiple mediation and a moderation effect. According to Hayes (2009), a mediation model assumes that a specific independent variable exerts an effect on a dependent variable through one or more mediating variables. In this research the hypothesis that the independent variable ‘SES’ influences the drivers of Healthy-Shopping Dynamics ‘Extremeness’ and ‘variability’ and purchase related variables such as ‘the percentage of healthy, unhealthy and discounted purchase decisions’, which in turn influence the dependent variable ‘end-of-trip basket health’, is tested. Furthermore, the hypothesis that the independents variable ‘health promotion’ influences end-of-trip basket health though the percentage of healthy and unhealthy purchase decisions is tested.
30 occurs when the independent variable has no effect on the dependent variable when the mediator is present between the both variables, partial mediation occurs when the independent variable has both a direct effect on the dependent variable and an indirect effect through the proposed mediator.
Figure 1: Representation of a simple mediation model from Hayes 2013.
A simple mediation effect is present in figure 1. The coefficients ai, bi, and c’, can be obtained by using least-squared regression. From these coefficients the total effects can be calculated by summing the direct effect (c’) and the indirect effect (ab), c = c’ + ab. In many situations the independent variable is dichotomous or continuous and the effect of x is than easily estimated in the figure above. In such situation the paths from figure 1 can be estimated with the following regression equation, wherein ai = a1 and bi = b1:
(1) BasHEaPD = b0 + c’ SES + b1 AVExtrPD + r, where (2) AVExtrPD = a0 + a1 SES + r
In this equation ai in figure 1 = a1 and bi = b1, furthermore b0 anda0 in this equation are the intercepts and r is the residual. This simple model can also be specified for the other mediators in the conceptual model. The independent variable x in this research is a multicategorical variable, consisting of the socioeconomic status of three groups (low, medium, high). Hayes & Preacher (2014) introduced the concepts of relative direct, indirect and total effects for multicategorical variables. The difficulty with categorical variables lies in the fact, that to represent the full effect of socioeconomic status with 3 levels / groups, 3 - 1 parameter estimates are needed. The consequence of this is that the parameter estimates and model fit statistics represent the information about how the specified groups differ from each other (Hayes & Preacher, 2014). Often researchers avoid this issue by aggregating groups or removing data to get a dichotomous independent variable, this is neither ideal or required according to Hayes & Preacher (2014).
There are several coding strategies that be used to represent the different groups and their effects. With ‘indicator coding’ one group is not explicitly coded, the two dummy variables in this research for SES are set to 0 for cases in that group. This groups functions as the reference group because the parameters in this research that represent the difference between the different group are relative to
X = Socioeconomic status (SES) M = Average extremeness change per purchase decision (AveExtrPD)
Y = End-of-trip basket health (BasHealthPD)
31 this reference group. In the research SES level 1 is indicated as the reference group. To measure the different paths in figure 1 for the mediator ‘AveExtrPD’ , where SES is treated as a multicategorical variable with 3 levels and other dependent variables in the research are included, the following regression equation can be specified, wherein ai = a1 and 2 and bi = b1:
(3) BasHEalthPD = b0 + c’1 SES1 + c’2 SES2 + b1 AVeExtrPD + b2 DWstat + b3HPromDum + B4 %HealthyPD + b5 %UnhealthyPD + b6 AvePricePD + b7 %DiscountedPD + b8
NrFoodPD + r,
where(4) AVeExtrPD = a0 + a1 SES1 + a2 SES2 + r
In this equation ai in figure 1 = a1 and a2 and bi in figure 1= b1, furthermore b0 anda0 in this equation are the intercepts and r is the residual. The other variables in the equation are explained as follows: SES 1 = a dichotomous (dummy) variable with a 0 for SES level 1 and 3 and a 1 for SES level 2
SES 2 = a dichotomous (dummy) variable with a 0 for SES level 1 and 2 and a 1 for SES level 3 AveExtrPD = the average extremeness in shifts of kilocalories between purchase decisions DWstat = The Durbin Watson statistic that measures the variability between purchase decisions HPromDum = a Dichotomous variable wherein 1 is representing a week wherein there was a health promotion campaign and 0 representing weeks in where there was no health promotion campaign. %HealthyPD = the percentage of purchase decisions wherein a consumers chose fruits, vegetables, or potatoes.
%UnhealthyPD = the percentage of purchase decisions wherein a consumers chose chocolate, cookies or candy.
AvePricePD = the average price of food products bought per consumer.
%DiscountedPD = the percentage of purchase decisions that involved a discount. NrFoodPD = the number of food purchase decisions
The mediation model captures the direct effect of SES on end-of-trip basket health by c’1 and c’2. As SES1 is a dummy for SES level 2 and SES2 is a dummy for SES level 3, the direct effects of SES on end-of-trip basket health are relative to SES level 1. The indirect effect through average extremeness change per purchase decision is captured by a1b1 a2b1. As a1 is a dummy for SES level 2 and a2 is a dummy for SES level 3, the indirect effects through average extremeness change per purchase decision are relative to SES Level 1.
32 in, is assumed to vary for different levels of SES. Specifically, a less negative effect of the ‘Variatie = Voordeelweken’ on end-of-trip basket health for higher levels of SES. According to Preacher (2007), expanding the simple mediation model can be done in several ways. The model expanded with the moderation effect is and all mediation effects represented by the following equation:
(5) BasHEaPD = b0 + c’1 SES1 + c’2 SES2 + b1 AveExtrPD + b2 DWstat + b3 HPromDum + b4
%HealthyPD + b5 %UnhealthyPD + b6 AvePricePD + b7 %DiscountedPD + b8 HPromDum * SES1 + b9 HPromDum * SES2 + b10 HPromDum * %HealthyPD + b11 HPromDum * %UnhealthyPD + b12 NrFoodPD + r, where
(6) AveExtrPD = a0 + a1 SES1 + a2 SES2 + r (7) DWstat = a0 + a1 SES1 + a2 SES2 + r (8) %HealthyPD = a0 + a1 SES1 + a2 SES2 + r (9) %UnhealthyPD = a0 + a1 SES1 + a2 SES2 + r (10) AvePricePD = a0 + a1 SES1 + a2 SES2 + r (11) %DiscountedPD = a0 + a1 SES1 + a2 SES2 + r
The complete model captures the direct effect of SES on end-of-trip basket health by c’1 and c’2. The indirect effect through average extremeness change per purchase decision is captured by a1b1 and a2b1. As SES1 is a dummy for SES level 2 and SES2 is a dummy for SES level three, the direct effects of SES on end-of-trip basket health and the indirect effects though through average extremeness change per purchase decision and the other mediators are relative to SES level 1. Thus, effects of SES level 2 compared to SES level 1 and effects of SES level 3 compared SES level one can be tested by using sequential dummy coding, wherein dummy variable 1 (SES1) has a value of 0 for SES level 1 and a value of 1 for SES level 2 and 3 and dummy variable 2 (SES2) has a value of 0 for SES level 1 and 2 and a 1 for SES level 3, the difference between SES level 2 and 3 can be obtained.
As can be seen in the equations above, 6 mediation models are required to test the effects of the different levels of SES through the mediators on end-of-trip basket health. To test hypothesis 10 and 11, two other mediation models are required. The equation to estimate the indirect effects of the health promotion variable through the percentage of healthy and unhealthy percentage of purchase decisions on end-of-trip basket health is specified below.
(12) BasHEaPD = b0 + c’HPromDum + b1 SES1 + b2 SES2 + b3 AveExtrPD + b4 DWstat + b3 HPromDum + b5 %HealthyPD + b6 %UnhealthyPD + b7 AvePricePD + b8 %DiscountedPD + b9 HPromDum * SES1 + b10 HPromDum * SES2 + b11 NrFoodPD + r, where
33 To test the specified hypothesis, the PROCESS-tool for SPSS for multiple moderated mediation analysis from Hayes (2013) is used. Hayes (2013) uses a bootstrapping method to test for mediation effects. This method is preferred over the causal steps method of Baron & Kenny (1986) as the conclusion is based on the indirect effects itself. Baron & Kenny (1986) assume that a mediation effects occurs when the effect of path c’ is less strong than the direct effect of the independent variable on the dependent variable. Several researches have shown that the causal steps approach is the least likely of several methods to detect the mediation effect (Fritz & Mackinnon, 2007). Furthermore, the Hayes (2009) method does not assume a normal distribution. The bootstrapping method is repeated 5000 in this research, as Hayes (2009) argues that 5000 times is sufficient to examine mediation effects.
