Hard discounters fighting their way into the sale of national brands – but how to conquer the customer’s heart?

Hard discounters fighting their way into the sale of

national brands – but how to conquer the customer’s

heart?

An investigation of the underlying reasons for a customer’s store choice

concerning the purchase of national brands.

2

Hard discounters fighting their way into the sale of

na-tional brands – but how to conquer the customer’s heart?

An investigation of the underlying reasons for a customer’s store choice

concerning the purchase of national brands.

Master thesis, MSc Marketing Intelligence

University of Groningen, Faculty of Economics and Business June 26, 2017

I.D. van Outersterp Tweede Willemstraat 27a

3

MANAGEMENT SUMMARY

Hard discounters are the fastest growing grocery format in Europe. Yet, to decrease their overreliance on price-based competition, many hard discounters increasingly follow an ex-pansion strategy wherein they add national brands to their private label-dominated assortment (Deleersnyder and Koll, 2012). As national brands were traditionally only sold at the service retailer, it raises important questions on the customer’s shopping behavior. Will customers stay loyal to the service retailer to purchase their national brands, or will they prefer visiting the hard discounter from now on? And more specifically, what are the underlying reasons or influences of this store choice? Consequently, this study is designed to examine where cus-tomers prefer to buy their national brands and the underlying reasons for this preference.

The research is based on data generated by an online survey among 420 Dutch cus-tomers. Goal of the survey is to measure, of each of the customers, the motivation behind his or her store choice when shopping for national brands. Based on the results, we can conclude that currently most of the customers still shop at the service retailer for their national brands. Furthermore, the attributes atmosphere, assortment, parking facilities, price, trip duration, price/quality ratio and accessibility are proven to be successful estimators of the customer’s store choice for national brands. In other words, the value a customer attaches to one of these store attributes determines how much a customer spends at the service retailer and the hard discounter. Specifically, the results indicate that customers who place most importance on atmosphere and price in their store choice mostly prefer service retailers. In contrast, custom-ers who emphasize accessibility, location and price/quality ratio represent the potential target market of the hard discounters. For parking, assortment and duration the findings are mixed among the customers, but for now it seems that most customers who attach value to parking are seconded to hard discounters as well. Lastly, the store attributes that play the largest role in a customer’s store choice are price/quality ratio and the trip duration.

The study contributes to the literature in that it creates insight on how a customer’s store choice is determined in a modern, dynamic retail environment, characterized by a battle between service retailers and hard discounters in the national brand market. An important implication to derive from the findings is that careful consideration of the customer will al-ways be of utmost relevance, as an understanding of the customer’s values and preferences enables the researcher to predict the customer’s store choice for his or her favorite national brand. Moreover, the findings support marketing practitioners in their awareness of the de-terminants of store choice. This way, brand manufacturers will be enabled to make a consid-erate decision on which retail format to use to sell their national brands and how to adjust their marketing mix to this decision. At the same time, the study provides guidelines to retail-ers on their potential target market and how to target these customretail-ers in order to generate a customer growth that has never been reached before.

4

PREFACE

To stay within the field of marketing, I would like to dedicate this master thesis to my beloved female friends that I made while being an active member at the Marketing Association of the University of Groningen. After spending long-lasting hours in the library together, alternated by the highly necessary social events, I witnessed the graduation of four passionate and hard-working women. Now all going by name of Master of Science Marketing Intelligence, it is an honor to be the final one and thereby to complete my thesis.

Furthermore, I would like to express my gratitude to dr. Erjen van Nierop, my supervisor, for his diligent guidance and relentless support throughout the progress of my thesis. His academ-ic insights, proposed refinements and professional suggestions have been important aspects in completing this thesis.

5

TABLE OF CONTENTS

MANAGEMENT SUMMARY ... 3

PREFACE ... 4

LIST OF TABLES AND FIGURES ... 6

ABBREVIATIONS ... 6

1 INTRODUCTION ... 7

2 THEORETICAL FRAMEWORK ... 9

2.1 Main actors in the retail market ... 9

2.1.1 Brand manufacturer ... 9

2.1.2 Service retailer ... 9

2.1.3 Hard discounter ... 9

2.2 National brand introductions at hard discounters ... 10

2.3 The customer ... 11 2.4 Store choice ... 12 2.4.1 Price ... 12 2.4.2 Quality of merchandise ... 13 2.4.3 Assortment ... 14 2.4.4 Atmosphere ... 15 2.4.5 Location ... 16 2.4.6 Parking facilities ... 17 2.4.7 Friendly personnel ... 17 2.5 Conceptual model ... 18 3 METHODOLOGY ... 19 3.1 Data collection ... 19 3.2 Research method ... 19

3.3 Measurement of the constructs ... 20

3.4 Plan of analysis ... 22

4 RESULTS ... 23

4.1 Descriptive statistics ... 23

4.2 Factor analysis ... 23

4.3 Hypotheses testing ... 24

4.3.1 Effects on current NB spending at Lidl ... 25

4.3.2 Effects on hypothetical NB spending at Lidl ... 30

5 DISCUSSION ... 35

5.1 Discussion of the findings ... 35

5.2 Theoretical implications ... 40

5.3 Practical implications ... 40

5.4 Limitations and suggestions for further research ... 41

6 REFERENCES ... 42

7 APPENDICES ... 50

7.1 Survey ... 50

6

LIST OF TABLES AND FIGURES Tables

TABLE 1 Summary of the means, standard deviations and correlations

of the DV and IVs 24

TABLE 2 Results binomial logistic regression analysis 27

TABLE 3 Results multiple regression analysis 29

TABLE 4 Summary results 34

TABLE 5 Segmentation of the models 63

TABLE 6 Regression estimates for ΔSOW 63

TABLE 7 Customer characteristics per segment 64

Figures

FIGURE 1 Conceptual model 18

7

1 INTRODUCTION

Hard discounter (HD) stores are the fastest growing grocery format in Europe (IGD Research, 2011). Examples of the format’s key food players in the Netherlands are Aldi and Lidl. HDs distinguish themselves by offering products at significantly lower prices than those offered by competing formats (Denstadli, Lines, & Grønhaug, 2005). To do so, they economize on fac-tors as customer service and store layout while focusing on low costs and high stock turnover (Gijsbrechts, Nisol, & Campo, 2008). Essential to the HD is a very narrow product assortment (Cleeren, Verboven, Dekimpe, & Gielens, 2010) and an emphasis on private-label (PL) prod-ucts, offering few, or even no, national brands (NB).

While many key features of HDs remain, in a number of areas there have been significant changes recently (IGD Research, 2011). This paper focuses on a recent trend: to decrease their overreliance on price-based competition, many HDs expand their assortment by adding (more) NBs to their PL-dominated assortment (Deleersnyder and Koll, 2012; Lourenço and Gijsbrechts, 2013). Examples of NBs that are now available at HDs are Unilever’s Unox and Pepsi Cola’s Lays (Distrifood, 2017a).

Deleersnyder, De Kimpe, Steenkamp and Koll (2007) were the first to study NB intro-ductions at HDs and the characteristics that make for their success. They demonstrate an in-crease in brand- and category sales for both the brand manufacturer (BM) and the service re-tailer (SR). Deleersnyder and Koll (2012) confirm this finding by demonstrating that the in-troductions enable the BMs to grow their total performance in excess of the market. Likewise, Steenkamp and Geyskens (2014) validate an increase in sales and customer traffic for the HD. Other scholars have contributed to the literature by taking different perspectives and outcome metrics. For instance, Lourenço and Gijsbrechts (2013) consider the implications for the HD in terms of store image and the chain’s share of wallet (SOW).

Taking these outcomes together, it seems that the potential benefits to the BM, SR and HD are a well-studied area. Yet, one player in the retail market has been relatively understud-ied: the customer. Little is known about how the entry of the HD in the NB market changes customer purchase behavior. How are shopping patterns affected by this trend in the market-place? Will customers stay loyal to the SR or will they buy their NBs at the HD from now on? Singh, Hansen and Blattberg (2006) and Vroegrijk, Gijsbrechts and Campo (2013) tried to answer these questions by analyzing HD market entries and the way customers reallocate their grocery purchases thereafter. Their findings show significant shifts in customer behav-ior, with changing store visits and basket sizes in particular. Deleersnyder and Koll (2012) contribute by assessing to what extent customers switch channels (existing SR vs. HD) once they find their branded offerings at the HD. Their results confirmed the study of Sloot, Fok and Verhoef (2006) in that customers are typically more loyal to their stores than to brands.

8 not taken into account how the process might differ in a dynamic retail environment charac-terized by NBs. This gap in the literature causes the author of this study to focus on drivers behind a customer’s store choice for NBs. Specifically, this research aims to contribute to the retail marketing literature by investigating where customers prefer to buy their NBs and the underlying reasons for this preference. The study is also relevant for managers as both brand manufacturers and retailers need to be aware of the determinants of store choice. This way, brand manufacturers will be enabled to make a considerate decision on which retail format to use to sell their national brands and how to adjust their marketing mix to this decision. At the same time, the study provides guidelines to retailers on their potential target market and how to target these customers in order to generate significant customer growth.

This leads to the following research question:

Why do customers prefer to shop where they do for national brands?

The study focuses on three research questions. First, where do people buy their NBs: at the SR or at the HD? Second, what store attributes drive the customers to this place? Third, what is the relative importance of the store attributes in this decision? The answers to these questions are studied by a two-part regression model and a latent class regression using a unique dataset of 420 Dutch customers, obtained specifically for this study.

9

2 THEORETICAL FRAMEWORK

This chapter demonstrates the relevant literature to the study. It starts with an introduction to the context by discussing the main actors in the market of NBs. This is followed by an analy-sis of how HDs expand their product assortment with NBs. Having discussed the different stores that sell NBs, the chapter continues by elaborating on the main subject: the customer’s store choice. Hypotheses are built by focusing on each determinant of store choice specifical-ly. The chapter concludes with a visual representation of the conceptual model.

2.1 Main actors in the retail market

2.1.1 Brand manufacturer

BMs are the creators of NBs: “brands created and owned by the producer of a product or ser-vice” (Kotler and Armstrong, 2015). The widely recognized trademark distinguishes itself from a regional or test-market brand through national distribution and advertising (Ostrow and Smith, 1988). They also differ from PLs (also called ‘retailer’s brand’) in that the latter are owned by retailers and distributed at their stores. BMs profit from NB buyers as they are characterized by a higher total spending (Ailawadi and Keller, 2004). Other benefits of selling NBs are a better price due to higher price elasticity, more bargaining power, brand loyalty retention and better control of distribution (Kotler and Armstrong, 2015). A disadvantage is the necessity for brand promotion to be successful, which is difficult for small manufacturers of unknown brands. Examples are Pepsi Cola and Head & Shoulders(De Standaard, 2005).

NBs are sold at various store formats (Levy, Weitz, & Grewal, 2014). González-Benito, Muñoz-Gallego and Kopalle (2005) define store formats to be broad, competing cate-gories that offer benefits to match the needs of various shopping situations and/or customers. Major retailer store formats are specialty stores, department stores, supermarkets, conven-ience stores, discounters, off-price retailers and superstores. Traditionally, most customers shop for NBs at the regular supermarket, also called the SR.

2.1.2 Service retailer

10

2.1.3 Hard discounter

HDs distinguish themselves from SRs by offering products at significantly lower prices than those offered by competing formats (Denstadli et al., 2005). The highly competitive price levels are the result of their explicit strategic focus (Lourenço and Gijsbrechts, 2013) on the use of minimal assortments and service, the implementation of a simplified, functional ‘no-frills’ store format, high stock turnover and little promotional activity (Deleersnyder et al., 2007). Meanwhile, they economize on factors as customer service and store size (500-1500 m2) (Gijsbrechts et al., 2008). This strategy translates into streamlined and cost-effective op-erations in the supply chain. An essential element to the HD is the emphasis on PL products, offering few, or even no, NBs (Cleeren et. al, 2010).

Discounters can be classified into hard and soft discounters (Denstadli et al., 2005; Colla, 2003) based on the number of products carried, operating costs, pricing levels and the role of PLs. HDs often carry a narrow product assortment (Cleeren et al., 2010), thereby limit-ing the range of product categories and stock keeplimit-ing units (SKU) per category, often far be-low the numbers typically offered in (big) supermarkets (Steenkamp and Kumar, 2009). This is at its most extreme among operators like Aldi, whose in-store product count can be as low as 900 (IGD Research, 2011). Both the operating costs and prices are very competitive, with 90% of the assortment accounting for exclusive PLs. Examples are Aldi or Lidl. Differently, soft discounters carry up to 3000 SKUs, with PLs making up to 50 percent of the assortment. Operating costs and price levels are generally higher. Nettorama and Dirk are seen as soft discounters. The remainder of this research focuses on the hard discounters. Interestingly, the HD format is the fastest growing format in grocery retailing in Europe (Deleersnyder and Koll, 2010). This remarkable success is mainly rooted in their ability to offer low prices (Lou-renço and Gijsbrechts, 2013).

2.2 National brand introductions at hard discounters

Despite their success, HDs realize that an overreliance on price-based strategies increases their vulnerability in the market. New ways of differentiation are therefore sought to stay competitive. Amongst others, Pan and Zinkhan (2006) argue that HDs should focus on a wide product assortment or selection, as assortment is the most influential factor to store choice. A highly recommended assortment extension strategy is the introduction of NBs to a PL-dominated assortment (Deleersnyder et al., 2012).The grocery-retailing landscape dramatical-ly as a result of this strategy. Traditionally, HDs were not deemed to list high-status NBs (Deleersnyder et al., 2007). Yet, in a UK survey, more than one third of the BMs have already reported to be approached by HDs to support a wider assortment (Gielens, Gijsbrechts, & Dekimpe, 2014). Now that NBs are sold at both the SR and the HD, formats increasingly compete with one another for the same customer (Cleeren, DeKimpe, & Verboven, 2006).

11 face the risk that the sale of NBs comes at the expense of sales at other retailers, particularly those with lower margins (Deleersnyder and Koll, 2012). That being said, SRs who already sell NBs can threaten to delist the NB that also is sold at the HD competitor (IGD, 2011).

Nonetheless, the assortment expansion also comes with many potential benefits. HDs use NBs as major store traffic builders, as customers might switch stores due to increased assortment attractiveness (Deleersnyder et. al, 2007; Dhar and Hoch, 1997). NBs are also a source of store loyalty and in turn higher profitability (Corstjens and Lal, 2000). The higher priced NB might also increase demand for PLs offered within the same store. That is, by pro-moting NBs, the HD differentiates its own brands and turns the PL price into the reference price of the category. The large price difference might then influence final brand selection (Oubiña, Rubio, & Yagüe, 2006). A mixed offering of NBs and PLs may thus enhance the HD’s performance and create a sustainable competitive advantage (Deleersnyder et al., 2007; Corstjens and Lal, 2000).

The collaboration is also valuable to the BM. It is an effective way to keep PLs in check, especially in categories with high variance in shares (Dhar and Hoch, 1997). Presence on the HD shelves also contributes to the alleviation of the dependency on SRs and their PLs (Ailawadi, Pauwels, & Steenkamp, 2008), as it puts constraints on NB margins to compete with the HD’s prosperity (Lourenço and Gijsbrechts, 2013). Besides, higher NB distribution proliferates market share (Reibstein and Farris, 1995) as the average sales elasticity is proven to be .74 for distribution breath (Ataman, van Heerde, & Mela, 2010). The BM has thus taken a superior position in where it sells NBs through two channels instead of one. Hereby, the BM hopes to capitalize on the growth of the HD channel, while simultaneously emphasizing the strength of selling through the SRs.

2.3 The customer

SRs and HDs emphasize different customer benefits and can therefore be seen as targeting different customer segments (Cleeren et al., 2010). Though, the rapid dynamics of retail forms has transformed not only the competitive structure of the industry, but also the way in which customers shop (Singh et al., 2006). The question arises how the customer responds to this new channeling strategy on NBs. Currently, around 80% of the customers visits multi-ple stores, with an average of 2.6 supermarket chains every month (EFMI Business School, 2014). The main reason why multiple-store shoppers traditionally shop at the SR is because it serves their need in categories in which they want the “best of the best (NBs)” (Nielsen, 2008). There is however a need for new insights in the customer store choice now that those best products (NBs) are sold at HDs as well. In particular, how are customers' shopping pat-terns affected by this trend? Do the customers stay loyal to the SR or will they buy their prod-ucts at the HD?

shop-12 pers who already visited other stores than the SR are most prone to switch stores. But, they also confirm customers to be more loyal to stores than to brands (Sloot et al., 2006; Vroegrijk et al. 2013) and that “cheap-and-lean” HDs will never fully replace SRs as the format of choice, even though they offer the same NBs. Instead, customer might choose to “trade down” in some store features and “trade up” in others by visiting both formats in parallel. HDs would thus turn into complements to SR rather than firm substitutes (Deleersnyder and Koll, 2012). Lastly, Singh et al. (2006) state that certain observed households characteristics, such as distance to the store, shopping behavior and product purchase behavior are useful in explaining a customer’s store choice.

2.4 Store choice

The research examples shed light on the fact that the range of general models to store choice drivers is extensive. The most widely accepted model is the NDB model (Uncles, Ehrenberg, & Hammond, 1995), which predicts the occurrence of store visits based on retailer size. Yet, this model is only one of many - most papers review different models and perspectives. There is no transparent conclusion on the drivers behind a customer’s store choice. The argument of Doyle and Fenwick (1974) that store choice is highly correlated to store image is nonetheless well supported. Based on this assumption, the factors that store image are used as a proxy for the factors that influence store choice in the remainder of this study. Existing literature takes a dominant attitudinal perspective that treats store image as a result of a multi-attribute model (Bloemer and De Ruyter, 1998) wherein image is stated as a function of salient store attrib-utes. Those attributes are then evaluated and weighted against each other. Store image is in the end expressed as the compound of a customer’s perceptions of a store on different salient attributes (Marks, 1976;James, Durand, & Draves, 1976).

Unfortunately, there is no universal consensus on the classification of store image ei-ther. This is largely due to its multidimensional structure. Plentiful studies have distinguished different store attributes that all contribute to the overall store image, and thereby to store choice (e.g. Kunkel and Berry, 1968; Marks, 1976). Doyle and Fenwick (1974) distinguish five elements: product, price, assortment, styling and location. Lindquist (1974) differentiates nine attributes: merchandise, clientele, convenience, service, promotion, physical facilities, store atmosphere, institutional factors and post-transaction satisfaction. One of the most cited authors in the store image literature is Bearden (1977). He considers price, quality of mer-chandise, assortment, atmosphere, location, parking facilities and friendly personnel to be most important in defining store choice. For the purpose of this study we use Bearden’s (1997) classification of store image attributes. The remainder of chapter 4 outlines these at-tributes and discusses how their importance to the customer affects customer’s store choice. Most often, a customers’ assessment on a store attribute influences a customer’s satisfaction. Customer satisfaction is in turn related to the choice of a specific store (Demirgünes, 2014).

2.4.1 Price

13 Customers develop general price perceptions in a store (Monroe and Lee, 1999). These perceptions are predominantly determined by high unit prices and high purchase frequency, complemented by price-related cues such as frequency of price advantages and non-price re-lated cues like quality levels (Brown, 1969; Desai and Talukdar, 2003). Although the retailer price format should logically play a large role here, customers may not always form valid perceptions of store prices (Dickson and Sawyer, 1990) due to subjective perceptions.

Research shows that the effect of store-level price promotions on store choice is sig-nificant (Mulhern and Leone, 1990), but weak (Volle, 2001). According to Blattberg, Briesch and Fox (1995), store promotions generate added traffic, leading to the purchase of comple-mentary products at full-margin. Yet, this has never been demonstrated. Others (e.g. Walters, 1988) argue price promotions to impact store switching only indirectly, as promotions only impact a customer’s category purchase decision when the customer is already visiting this store. This research stream is followed by Galata, Randolph, Bucklin & Hanssens (1999), who argue that price format loyalty rarely causes customers to switch towards other formats.

Most importantly, managers use the retail price format to increase store traffic, shop-ping trip frequency and customer spending (Bell et al., 2001). Retailers select price formats of a continuum secured by High-Low Promotional Pricing (HILO) at one end and Every Day Low Pricing (EDLP) at the other (Ailawadi and Keller, 2004). HILO involves setting higher average prices (Olbrich, Jansen, & Hundt, 2017), but with a high variability between regular and promoted prices. The frequent promotional prices are used to attract small basket shop-pers that search for occasional price deals on individual products. SRs often use this format. Contrarily, the EDLP format involves stable but low prices, thereby avoiding price promo-tions. This format is used by HDs and attracts price-sensitive, large basket shoppers unwilling to spend effort in finding the best price for each individual product (Cleeren et al., 2010). Alt-hough there is no dominant format, it is important to realize that customer segments differ in their perception and response to dissimilar formats (Ho et al., 1998). Different formats lead to customer heterogeneity (Rhee and Bell, 2002) and to inherently similar types of customers showing altered behavior (Tang et al., 2001).

To summarize, SRs follow a HILO strategy targeted on small basket shoppers. At the same time, HDs pursue the EDLP strategy, thereby meeting the needs of large basket, price-sensitive shoppers. Therefore, it is assumed that customers who consider price as an important store attribute, prefer the lower-priced HD to the higher-priced SR. This lead to the first hy-pothesis:

Hypothesis 1: The importance of price to a customer will be positively related to the choice of HD stores and negatively related to the choice of SR stores.

2.4.2 Quality of merchandise

14 elements of the merchandise. Both tangible and intangible characteristics are taken into ac-count, just as performance, reliability, features, serviceability and conformance. Notably, Zeithaml (1988) emphasizes that perceived quality is different from objective or actual quali-ty. Whereas objective quality refers to verifiable and quantifiable superiority on some prede-termined standards (Curry and Faulds, 1986), the personal judgment is generally made on the products that exists within a customer’s evoked choice set (Maynes, 1976). This evaluation takes place in a comparison context, where the product is being evaluated against a set of oth-er products. Both intrinsic cues (physical product charactoth-eristics, including size, color, tex-ture) and extrinsic cues (external product characteristics like price, brand image, level of ad-vertising) play a role in the perception (Zeithaml, 1988). Perceived quality may eventually lead to customer satisfaction and in turn to the choice of a specific store.

Merchandise quality has often been defined as a component of store image (e.g. Lind-quist, 1974). Yet, this study proposes merchandise quality as an antecedent of store image. This view aligns with Mazursky and Jacoby’s (1986) definition of store image as “an af-fect/cognition that is inferred from a set of perceptions”. They observe the influence of mer-chandise quality perceptions on retail store image and demonstrate that customers base their quality perception on the brand names carried by the retailer. Several authors (De Wulf, Schröder, Goedertier, & Van Ossel, 2005; Walker, 2006) verify that customers perceive the quality of NBs to be of superior quality compared to that of PLs. Following this reasoning, the SR is perceived to have a higher quality of merchandise because its assortment is more NB-dominated. Therefore, it is assumed that customers who consider quality of merchandise to be an important store attribute, choose the higher-quality SR over the lower-quality HD. An im-portant note is that the addition of NBs to the HD-assortment is expected to have a positive impact on the perceived merchandise quality of the HD (Bronnenberg and Wathieu, 1996), since the NBs are likely to improve the attractiveness of the HD’s assortment. Yet, the SR will still be characterized by an assortment that is much more NB-dominated than the assort-ment of the HD and therefore be perceived as of higher quality. It is thus hypothesized that:

Hypothesis 2:The importance of quality of merchandise to a customer will be positively relat-ed to the choice of SR stores and negatively relatrelat-ed to the choice of HD stores.

2.4.3 Assortment

The third factor influencing store choice is assortment. Officially defined as “the number of different items in a merchandise category” (Levy et al., 2014: 245), it is debated to be more important than retail prices in store choices (Briesch, Chintagunta, & Fox, 2009). The key challenge for food retailers is thus to offer a unique product assortment that is unavailable at competing retailers and that attracts customers (Simonson, 1999). Product assortment general-ly holds the number of product categories offered (width), the number of brands within a product category (length) and the number of SKUs within a product category (depth). Though, other factors such as the size per brand, proportion of unique SKUs and the presence of the households’ favorite brands also fairly influence store choice (Briesch et al., 2009).

15 more variety as it meets the demands of a segmented market. Further, more choice not only involves customers that embrace variety seeking, but it also drives intrinsic motivation, life satisfaction and perceived control (Narasimhan, Neslin, & Sen, 1996; Van Trijp, Hoyer, & Inman, 1996; Langer, 1983). There are, however, two sides to the assortment model. The “more is better” assumption is criticized (Broniarczyk and Hoyer, 2006)as the role of assort-ment changes during the customer journey. While customer attraction might be higher with an extended assortment, conversion (purchase intention) might be higher in case of a limited offering (Iyengar and Lepper, 2000). Too much choice is associated with a high cognitive load for customers, leading to regret or even avoidance of a choice (Schwartz, 2004) and ul-timately to a decrease in satisfaction. Along this line, retailers that offer fewer SKUs per brand, fewer sizes per brand and fewer unique items also increase their opportunity to be pre-ferred (Briesch et al., 2009). Designing the optimal assortment thus holds a trade-of between the benefits and the costs of various assortment sizes.

As store choice is only about customer attraction, this study follows the reasoning that an increased (perceived) assortment size increases the likelihood of the store to be the pre-ferred choice (Ailawadi and Keller, 2004). One should know that a SR carries a wide and deep assortment, whilst a HD is characterized by a narrow assortment. A SR would thus be more attractive in terms of assortment than the HD. The reader should keep in mind though that a larger number of SKUs does not directly translate into better customer perceptions – a difference in number of brands offered also plays a role. Whilst a SR carries a large propor-tion of NBs, HDs offer few or sometimes even no manufacturers brands. The SR is therefore more likely to capture a customer’s preferred brand. Altogether, it is suggested that customers who attach value to the store’s assortment, prefer the SR to the HD. Therefore, the third hy-pothesis is:

Hypothesis 3: The importance of assortment to a customer will be positively related to the choice of SR stores and negatively related to the choice of HD stores.

2.4.4 Atmosphere

Store atmosphere (also called: store environment or design) has a substantial influence on customer store choice behavior. It refers to the part of the customer journey where customers observe physical characteristics and surroundings when interacting with any part of the retail-er (Stein and Ramaseshan, 2016;Kotler, 1973). Retailers wish to create a unique store experi-ence that moves the target market to purchase (Kotler and Armstrong, 2015). From the envi-ronmental psychology perspective, store atmosphere is important because effective environ-mental stimuli will instigate certain emotional states that in turn lead to preferred customer responses, e.g. store patronage or repeat shopping (Donovan and Rossiter, 1982).

16 hold the other people (other customers, service personnel) in the environment, both in terms of numbers as in appearance and behavior. Design factors exist at the forefront of our aware-ness. They are either aesthetic (scale, architecture) or functional (layout, comfort). Aesthetic or functional cues in the retail store environment provide the customer with information about the quality, capabilities and image of the store (Bitner, 1992).

Previous literature presents specific environmental cues that fall within the three di-mensions and that are useful to distinguish between a SR and a HD. Regarding store ambi-ence, Garlin and Owen (2006) identify the presence of music to positively affect shopper pat-ronage behavior, especially with familiar and likable music. SRs often play music in their store, whereas the “no-frills” HD format does not allow for music to be played. In terms of design factors, Bellizzi and Hite (1992) indicate a store design cue in that they found the color blue to make customers more relaxed and therefore spend more time in-store browsing than compared with other colors. This in turn promotes purchase intentions of an increased number of items. Summers and Hebert (2001) discuss a similar effect as a result of a 500W lightning in the store. Here, note that SRs are characterized by more cozy light, whilst a more “cheap”, bright light in the store portrays the “no-frills” HD format. Lastly, for a discussion on social factors please refer to 2.4.7. To summarize, the “no-frills” format characterizes HDs to have a cleaner store atmosphere than SRs. HDs put less emphasis on the non-product related attrib-utes of the store, even though it has been confirmed that non-product related attribattrib-utes of the retail store environment significantly influence store image, attitudes and store loyalty (Bit-ner, 1992). The atmospheric profiles of SRs are therefore better perceived by customers in general. This leads us to the following hypothesis:

Hypothesis 4: The importance of atmosphere to a customer will be positively related to the choice of SR stores and negatively related to the choice of HD stores.

2.4.5 Location

Store location and the distance to travel to a store have gained significant research attention. From the beginning, authors like Meyer and Eagle (1982) argued location to be amongst the most important drivers of store choice. The main reason for this is its function in a customer’s assessment of total shopping costs, especially for retailers whose SOW comprises small bas-ket shoppers and fill-in trips. Large-scale retailers hence focused largely on their geographic positioning strategy. Despite this, nowadays location no longer explains a major portion of variance in store choice because of new warehouse retail formats, online distribution channels and greater driving distances (Ailawadi and Keller, 2004; Ho, Tang, & Bell, 1998).

17 likely to attract customers from greater distances (Graig, Ghosh, & McLafferty, 1984). Unfor-tunately not all retail formats (convenience, drug, supermarket) have the flexibility in select-ing a location. Lastly, parkselect-ing facilities is likely to be covered under convenience as well, but as we follow Lindquist (1974)’s theory, this concept is discussed separately in 2.4.6.

SRs offer a wide range of choice to trolley-stacking weekly shoppers. They search for more distant locations as a larger format suffers from more restrictive criteria on its location selection. Luckily, a larger assortment also increases a customer’s willingness to bridge larger distances. Contrarily, the location strategy of HDs is one of proximity retailing (Zentes, Mor-schett, & Schramm-Klein, 2016). HD stores often squeeze themselves into easily accessible cost- and traffic-oriented sites with a focus on little tenure costs, e.g. community locations or periphery sites. They head towards a strategy in where they aim at frequent, small basket shoppers (Shadbolt, 2015). The centrally located stores are ideal to reduce customers’ transac-tion costs of time and effort. HDs thus better match the need of the customer who perceives location as an important attribute for store selection. As such, the following is hypothesized:

Hypothesis 5: The importance of location to a customer will be positively related to the choice of HD stores and negatively related to the choice of SR stores.

2.4.6 Parking facilities

Parking facilities form an important factor affecting store choice. Their availability is directly linked to store location and availability, and thereby to store convenience (Finne and Sivonen, 2008). Parking services provide a competitive advantage (Ok Kim and Jin, 2001) in the form of a good shopping experience (Bhukya and Singh, 2016). Facilities like ample parking and long operating hours are used to increase store traffic (Hansen and Deutscher, 1978). Custom-ers value covered parking spots as well as they prefer going to the store and back to the car without getting themselves (or their children) wet (Finne and Sivonen, 2008).

SRs generally have sufficient parking facilities. This is in all probability linked to store location: if the store is located on a distance and mostly reached by car, it is basically a requisite to provide a sufficient number of parking spaces. This is less the case with HDs as they are argued to be located more closely within the city center. It should be noted that new HDs have shown to be more oriented towards the upper end of the traditional size band, in-cluding plentiful car parking space (IGD research, 2011). Yet, for the purpose of this study HDs are assumed to have less parking facilities because of their central location. Customers who perceive parking facilities to be important in their store choice might thus be better of going to the SR. The following is hypothesized:

Hypothesis 6: The importance of parking facilities to a customer will be positively related to the choice of SR stores and negatively related to the choice of HD stores.

2.4.7 Friendly personnel

18 moment of employee-customer interaction: the direct and indirect interactions customers have with employees when interacting with any part of the retailer (Stein and Ramaseshan, 2016). This interaction results in customer satisfaction towards the store. As friendliness is closely associated with service quality, it should be noted that only the perceived judgment on friend-liness matters. This judgment results from an evaluation process where customers compare their expectations with the actual service they perceive to have experienced (Grönroos, 1984; Baker et al., 2002). Darian, Tucci and Wiman (2001) argue that customers expect a certain threshold level of friendliness, after which there does not seem to be a large effect. Only when the threshold is not achieved, friendliness gets important and customers perceive to be treated unfriendly. To avoid a similar situation, it is recommend for a store’s personnel to receive training with the objective to achieve the desired threshold. It is important for personnel to maintain high service standards, especially since customers identify and associate the traits of service personnel with the stores they work for (Hollensen, 2011).

HDs largely neglect the possibility to offer vocationally trained personnel – or offer no (personal) service at all (Wortmann, 2004). SRs, on the other hand, may emphasize the level of customer service as an important part of the total shopping experience. Consequently, this study highlights that customers who consider friendliness of personnel to be an important store attribute most likely prefer the SR to the HD. This leads to the following hypothesis:

Hypothesis 7: The importance of friendly personnel to a customer will be positively related to the choice of SR stores and negatively related to the choice of HD stores.

2.5 Conceptual model

The literature review conceptualizes that shoppers base their store choice on seven factors: price, quality of merchandise, assortment, atmosphere, location, parking facilities and friendly personnel. The following figure shows the conceptual model corresponding to these factors, hypothesizing that all factors significantly relate to a customer’s store choice.

19

3 METHODOLOGY

This chapter outlines the research methodology. It starts with a description of the data collec-tion and the research method. This is followed by a discussion of the measurement of the con-structs and the plan of analysis.

3.1 Data collection

The data is collected by means of an online survey. The target group consists of Dutch cus-tomers, since the research is focused on supermarkets in the Netherlands. Before distributing the final survey, a pre-test was done among a small group of respondents with the objective of eliminating potential problems (Malhotra, 2010). The survey was sent out using convenience sampling and snowball sampling. First, subjects familiar to the researcher were contacted by e-mail, social media and personal contact. Subsequently, respondents were asked to distribute the survey to other acquaintances. Participants were allowed to complete the survey in their own time and anonymously. A chance to win a supermarket gift card worth €20,- was offered to motivate respondents to complete the survey. The combination of sampling methods ulti-mately generated a totalof 654 respondents that filled in the survey with a completion rate of 73%. Some observations had to be deleted, as participants did not complete the entire survey and/or did not take on the role of buying the bulk of the household’s groceries. A duration tracker was added so to track and manage the time respondents spent on the survey. Respond-ents spent 11 minutes on the survey on average. The study has been conducted in a length of two weeks (April 21 to May 4, 2017). The final sample consists of 420 observations. This number is deemed sufficient as factor analysis requires to have at least five times as many observations as there are variables (35 in this case).

3.2 Research method

Goal of the survey is to measure, of each of the respondents, the motivation behind his or her store choice when shopping for NBs. The final survey can be found in the appendix. The sur-vey is created and distributed through an online program called Qualtrics. The language used in the survey is Dutch, the native language of the country of interest (The Netherlands). Over-all, the survey contains 15questions and 24 factors to rate. The survey is divided into six parts and structured as follows:

First, the respondent has to agree with the general terms and conditions. These also in-clude information on the expectations of the respondent and the survey length. The respond-ent is also asked to read an explanation of the constructs of NBs and PLs. Goal of this survey section is to ensure that the respondent correctly understands and interprets the survey.

The second part concerns introductory questions on grocery shopping. Respondents who indicate that they are usually not the person doing grocery shopping for their household are friendly requested to exit the survey, as these answers are not deemed valuable to this study. Other respondents just continue the survey. The objective of this part is to increase the usefulness of the sample.

varia-20 ble (on a scale of €0 to €100) is used in order to maximize the amount of information collect-ed. The objective of this section is to gather information on store choice, in particular on the amount spent at Lidl (the HD that is used as dependent variable in our model). Lidl is chosen as the supermarket in question because when differentiating between Aldi and Lidl, Aldi al-ready has a considerate number of NBs in stock, whereas Lidl is relatively new in this field. Lidl therefore offers more room to experiment in relation to reality.

The fourth part of the survey contains an experimental set-up in order to collect more information on the store choice attributes store choice (the independent variables in our mod-el). This section is valuable to the study as it is expected that section 3 and 5 of the survey only contain a few informative observations. Reason for this is the assumption that only a few respondents currently shop at HDs to buy NBs. This section therefore incorporates built-in scenarios in order to measure the amount a respondent would spend on NBs at Lidl under certain hypothesized circumstances. Again a SOW variable is used to measure the amount spent. A 2x3 design (assortment x price) leads to a total of six different experimental condi-tions. On the one hand, each scenario includes either five or ten NBs. These new items are based on the NB top 100 2017 (Rensen, 2017). Yet, as the top 10 contains many overlapping product categories, products slightly higher listed are also used in order to avoid a bias to-wards particular product categories. The eventual product mix consists of ten unique items from different product categories. On the other hand, each scenario includes a 0%, 5% or 10% price discount relative to the regular prices in the SR Albert Heijn. All six scenarios are ran-domly assigned to the respondents.

In the fifth section the respondent is asked to evaluate 24 items on their importance in the choice of supermarket. The items are more elaborately discussed in section 3.3. Goal of this part is to collect data on the seven store attributes (the independent variables in our mod-el) in order to test the hypotheses.

The final part of the survey includes questions that together outline the demographic profile of the respondent, including the possibility to leave their e-mail address in order to have a chance in winning a supermarket voucher. After completing the survey, the respond-ents are thanked for their participation. The objective of this part is to gain additional insights into the sample so to be able to segment respondents and check for the representativeness of the sample. Another reason for inclusion is that socio-demographic variables might influence buying behavior as such (Gilbert and Jackaria, 2002).

3.3 Measurement of the constructs

The variables are measured using developed instruments and scales from existing literature. This subsection presents each variable and the associated measurement instrument.

Store choice. The dependent variable (DV) is officially defined as the store where

21

Store attributes. The theoretical framework argues seven attributes to influence store

choice. These attributes are our independent variables (IV). To investigate the importance of each attribute to the customer, the survey follows Fischer, Völckner and Sattler (2010) in that it uses a Likert scale in which respondents are asked to rate the importance of each attribute in their choice of supermarket. A 7-point scale (1: very unimportant, 7: very important) is used to collect most information and to make it easy for respondents to understand all questions (Malhotra, 2010).Multiple items are included per attribute, leading to 24 items in total. Price is operationalized as the amount a customer has to pay to buy a product. It is measured by the following items: low prices, appealing promotions, price/quality ratio. Quality of merchandise

assortment is operationalized as the perceived quality of the product assortment. It is

meas-ured by the following items: product quality, reliable product quality, products are in stock.

Assortment is operationalized as the products sold by the retailer. It is measured by the

fol-lowing items: new products, large assortment, and fresh products. Atmosphere is operational-ized as the store environment. It is measured by the following items: store layout, store hy-giene, space within the store, store is well ordered, opening hours. Location is operationalized as the place where the store is situated. It is measured by the following items: other stores close to the supermarket, store is nearby, online store, delivery services, and store accessibil-ity. Parking is operationalized as the facilities to deposit a car a the store. It is measured by two items: number of parking spaces and covered parking. Friendly personnel is operational-ized as the behavior of store personnel. It is measured by the following items: speed at the check out, competent personnel, friendly personnel. All items are randomly assigned to re-spondents and based on two sources. The first source is the Customer Relevancy Model by Crawford and Mathews (2001). They operationalize the constructs of product, price, access, service and experience into useful statements about store choice. Second, the items are based on the questions asked in the EFMI Shopper Monitor (EFMI Business School, 2015). The EFMI Business School, an academic knowledge institute in the Dutch food industry, devel-oped this survey to follow developments on the demand-side of the grocery market.

Price and assortment. The attributes price and assortment are also used in order to

cre-ate the scenario variables in survey section 4. The scenario variables are measured by a SOW variable. At the start of the survey, the respondent is asked how much he or she usually spends on NBs at the selected supermarkets. After showing each new scenario, the SOW question is again displayed to the respondent. This way, it is checked whether a change in any of the two attributes leads to an change in the amount spend at a HD (Lidl in this case).

Socio-demographic variables. The scales of the demographic variables are based on

the EFMI Shopper Monitor (EFMI Business School, 2015). The following are included: • Gender: Gender is operationalized as the sex of the respondent. Measured as a

nomi-nal variable with the answer options ‘male’ or ‘female’.

• Age: A respondent’s age, expressed in years. This interval variable has a range of val-ues between 16 and 83 years old.

educa-22 tion (MBO)’, ‘school for higher general secondary education or pre-university educa-tion (HAVO/VWO)’, ‘university of applied sciences (HBO)’ to ‘university (WO)’. • Type of residency: Operationalized as the current living situation of the respondent.

Measured by a 5-point scaled nominal variable, divided into the categories ‘living alone’, ‘with friends/students’, ‘with partner’, ‘with children’ and ‘other’.

• Place of residency: Operationalized as the home town/region of the respondent. Measured by a 7-point scaled nominal variable, expressed in the categories ‘Amster-dam’, ‘Rotter‘Amster-dam’, ‘The Hague’, ‘West’, ‘North’, ‘East’, and ‘South’.

• Work situation: Operationalized as the daily pursuits of the respondent. Measured by a 6-point scaled nominal variable, divided into the categories ‘full-time’, ‘part-time’, ‘retired’, ‘student’, ‘unemployed’, and ‘other’.

• Income level: Operationalized as the respondents’ gross monthly earnings. Measured by an 8-point scaled ordinal variable, with the options: ‘<1000’, ‘1000-2000’, ‘2000-3000’, ‘3000-4000’, ‘4000-5000’, ‘5000-6000’, ‘>6000’, ‘other/don’t want to say’

Other: The number of supermarket chains visited per shopping trip, the number of

shopping trips each week and the estimation of weekly spending on grocery shopping are all open questions that are answered on an interval scale. The answers to these questions provide the author with a more complete image of the sample.

3.4 Plan of analysis

The software programs used to analyze the data are SPSS 21 and Latent Gold 5.4. The follow-ing tests are conducted: a factor analysis, a binomial logistic regression analysis, a multiple regression analysis and a latent class regression analysis respectively.

Factor analysis. The survey includes 24 items that indirectly measure the attributes. The

goal of the factor analysis is to combine multiple of these items into one factor that represents a singular construct. That way, the number of original items is reduced into a smaller set of new, composite factors with a minimum loss of information. Items are grouped that correlate too strongly with each other. The resulting factors are used in subsequent analysis.

Binomial logistic regression and multiple regression analysis. A two-part model is

car-ried out for hypotheses testing. First, a binomial logistic regression is used to predict the like-lihood of the respondent shopping at Lidl. Second, a multiple regression analysis is conducted wherein the factors are regressed on the current SOW at Lidl. The predictor variables are checked for their significance to the DV. So far, all tests are performed using SPSS.

Latent class regression analysis (LCA). LCA is performed in order to get a better model

23

4 RESULTS

4.1 Descriptive statistics

The sample (N=420) consists of 112 (27%) men and 308 (73%) women. Most respondents (44%) are between 21 and 26 years, with the oldest respondent being 83. In terms of employ-ment, 65% of the shoppers work at least part-time, whereas 17% is still studying. Household sizes diverge largely, yet most respondents (57%) live together with friends or with their part-ner. The majority of the respondents (67%) has a gross monthly income between €0 and €3000. The sample is weighted to correct for the disproportional sample sizes in comparison to the population as reported by Centraal Bureau voor Statistiek (CBS) (2017) and to adjust the collected data to the target population from which the sample was drawn. The weighted variables are region and education level. After weighing, 35% lives in the West, 18% in the North and 39% in the Eastern part of the Netherlands. Almost everyone (95%) completed at least vocational post-secondary education (MBO), with the majority (50)% having attained university (of applied sciences). It is acknowledged that both the Southern part of the Nether-lands and the lower educated part of the population are still considerately underrepresented (8% and 5%). Reason for this is that they only contain 32 and 22 observations respectively. Weighting these observations would make statistical inference over-optimistic, as the sam-pling properties of the statistics would be driven by a trivial number of observations. Section 5.4 elaborates more on this. Two other figures that somewhat differ in comparison to the pop-ulation are gender and age. Yet, these can be logically explained. First, the deviation regard-ing gender distribution is explained in that still relatively more primary shoppers are women. Moreover, the underrepresentation of below 18 years old is explained by the fact that this group still lives at their parents and that one of their parents is often the primary shopper.

By comparing the sample with the results of the EFMI consumer trends report (2014), it is concluded that the sample sufficiently represents the average shopper. The respondents claim four supermarket chains to be within easy reach, from which they visited about two-third in the last month. They venture out for an average of 3.7 shopping trips each week and visit 1-2 supermarkets during each of those trips. One should only be aware that the number of average shopping trips each week is slightly higher (3.7 vs. 2.7).Lastly, the shoppers spend €75 on average on groceries each week, from which 41% goes to NBs and 59% to PLs.

4.2 Factor analysis

24

1. In-store atmosphere (Cronbach’s alpha = .789). Five items that reflect the extent to which

the respondents consider the in-store atmosphere to be important in their store choice. These items concern friendly personnel, competent personnel, space within the store, store hygiene and store layout respectively.

2. Trip duration(Cronbach’s alpha = .689). Measured by five items: speed at the check out, store is well ordered, products are in stock, opening hours, and store is nearby.

3. Assortment (Cronbach’s alpha = .817). Computed by three items: large assortment, new

products, and fresh products.

4. Price/quality ratio (Cronbach’s alpha = .714). Consists of five items: low prices,

price/quality ratio, product quality, reliable product quality and appealing promotions.

5. Parking (Cronbach’s alpha = .964). Measured by two items: number of parking spaces and

covered parking.

6. Accessibility (Cronbach’s alpha = .703). Computed by three items: delivery services,

online store, and store accessibility.

Please note that the item “other stores close to the supermarket” is left out of the analy-sis due to low correlation to the other items. The newly constructed factors are used in subse-quent analyses. This has implications for the hypotheses under study, as these factors do not allow all previously constructed hypotheses to be tested. Specifically, H3, H4 and H6 are still appropriate for investigation and remain similar. H1 and H2 are not investigated directly an-ymore, but combined into one factor called price/quality ratio. The factor analysis fully elimi-nates the possibility to investigate H5 and H7. Instead, the influence of trip duration and store accessibility is studied in the remainder of this research.

4.3 Hypotheses testing

TABLE 1 presents the descriptive statistics for the sample (N = 420). SOW Lidl and ΔSOW Lidl represent our DV for different analyses and are therefore both included. Note that the descriptives for SOW Lidl are given twice, once for the reduced sample (N = 53) and once for the full sample. The reasoning behind the reduced sample is explained in section 4.3.1.

TABLE 1

Summary of the means, standard deviations and correlations of the DV and IVs

Variables Mean S.D. 2 3 4 5 6 7 8 9 1. SOW Lidl > 0 (N=53) 68.04 14.65 2. SOW Lidl ≥ 0 (N=420) 3.00 12.56 1 3. ΔSOW Lidl 38.63 33.03 .15** 1 4. Atmosphere 4.90 1.05 -.20 .03 1 5. Duration 5.53 .83 -.30 .07** .36** 1 6. Assortment 4.76 1.38 .01 .09** .41** .24** 1 7. Price/quality 5.54 .89 .08 .24** .06 .29** .14** 1 8. Parking 3.52 1.99 .13* .09** .37** .15** .11* .13* 1 9. Accessibility 5.41 1.03 .04 .03** .15** .37** .14* .11* .61 1

25 TABLE 1 shows that many variables are correlated to each other. Notably there is a significant positive relationship between SOW Lidl ≥ 0and parking, r = .13, p < 0.05. For ΔSOW Lidl, all variables except for atmosphere show a significant correlation (p < 0.05). The IVs also highly correlate. Atmosphere is significantly positively related to duration (r = .36), assortment (r = .41), parking (r = .37) and accessibility (r = .15). Further, an increase in im-portance of duration leads to an increase in the imim-portance of assortment (r = .24), price/quality (r = .29), parking (r = .15) and accessibility (r = .37). In the same way, an in-crease in assortment positively influences price/quality (r = .14). A rule of thumb for interpre-tation of the correlations holds that correlations above >.5 show strong correlations. In our case, this is only parking and accessibility (.61). All other correlations are < .5 and thus show low to no significant correlation. To check for the reliability of the results, a collinearity diag-nostic test is performed on all IVs. As all tolerance factors > 0.2 and VIF < 5, there is no rea-son to assume multicollinearity.

4.3.1 Effects on current NB spending at Lidl

In order to analyze whether the factors in-store atmosphere, duration, assortment, price/quality, parking and accessibility influence store choice, the set-up is to perform a regu-lar multiple regression analysis with these factors regressed on SOW at Lidl. However, the database only contains a few informative cases and a very large number of non-respondent (or non-informative) observations, because many respondents currently do not shop at Lidl. That means, our variable SOW is heavy censored in that spending on NBs occurs at only 12.6% (53 out of 420) of the cases. The data thus contains a considerate number of zeros. In the rest of this analysis these 53 cases are further referred to as customers, and the rest as

non-customers. The problem with this type of dataset is that a thorough panel analysis is not

pos-sible. The solution, although not strictly correct, is based on a two-part model that has shown to be particularly relevant for marketing research with a large full sample that contains a very small number of respondents (Cramer, Frances, & Slagter, 1999).In this model, the customers are treated as if they constitute one large sample. The first step in the model is a binomial lo-gistic regression for the occurrence of shopping at Lidl. After that, a type I censored regres-sion model is performed for the amount of money involved when shopping at Lidl.

Logit analysis. First, a binomial logistic regression is conducted in order to estimate

26 very small and therefore considered negligible. Henceforth the remainder of this research will focus only on one reduced sample instead of discussing many, almost similar samples.

All the assumptions for the binomial logistic regression model are met in advance (Garson, 2009). To illustrate, the DV is a dichotomous variable with mutually exclusive and collectively exhaustive categories wherein the independent observations are given a value of 1 if they currently shop at Lidl and a value of 0 if they do not. The sample size is sufficient in that ML needs at least 15 cases per IV. In this case, 6 IVs means that 90 cases hold the mini-mum, being < 420. The model includes multiple continuous (interval) IVs that do not show multicollinearity (see 4.1). The Hosmer and Lemeshow test (p = .309) shows that the model is not a poor fit. Linearity of the continuous variables with respect to the logit variable is as-sessed via the Box-Tidwell procedure (Box and Cox, 1964). A Bonferroni correction is ap-plied using all twelve terms in the model resulting in statistical significance being accepted when p < .004 (Tabachnick & Fidell, 2007), resulting in all continuous IVs being linearly related to the logit of the DV.Only the assumption on outliers is not met as SPSS recognizes almost all cases with 1 as DV as a studentized residual with a value of >2.5 standard devia-tions. Still, these cases are kept in the analysis as they are necessary to conduct the regression.

The logistic regression model is expressed as follows:

P(Shopping at Lidl) = 1/(1+e-Zi) (4.1)

In where Z is a linear combination of the IVs and coefficients to be estimated:

Z =

B

+ B

X

+ B

X

+ B

X

+ B

X

+ B

X

+ B

X

+ ε

Where P is the DV, B0 is the intercept, Bn are the slope coefficients, ε is the error term, and “i”

refers to the ith respondent of the 420 respondents. Xn are the IVs (X1 = Atmosphere, X2 =

Du-ration, X3 = Assortment, X4 = Price/quality, X5 = Parking, X6 = Accessibility).

The results are shown in TABLE 2. The table shows the parameter estimates for a bi-nary logistic regression model for both the full sample (model A) and for the reduced sample (model B). Model A is statistically non-significant, χ2_{(6) = 8.495, p > .05. We can conclude}

27 The binary logistic regression on the full sample is insignificant due to the large num-ber of non-customers in the sample. ML is therefore also estimated for the reduced sample K = 2. This adjustment is appropriate as with binomial logistic regression it is allowed to use selective sampling as long as the constant is appropriately adjusted. Note however that the intercept is not adjusted in this study, as this is deemed unnecessary due to the explanatory nature of the IVs. The sample is reduced to N = 106 using stratified random sampling. The binomial logistic regression model with the reduced sample (K=2) is statistically significant, χ2(6) = 10.920, p < .05 (see TABLE 2). We can thus conclude that there is a significant rela-tion between the IVs and the DVs. This also indicates that the joint significance of all factors is not redundant and the model is better than the naive intercept model. The model explains 12.9% (Nagelkerke R2) of the variance in store choice, which is 9.3% more than model A. Yet, now only 66.6% of the cases are classified. This seems much lower compared to model A, but one should note that it is more difficult to correctly predict the outcomes for the re-duced sample than it is for the full sample. Simply said, in model A the chances of a value of 0 are of such a large extent that even when all values are predicted to be a 0, the classification rate would still be very high. The classification rate should therefore be taken into careful consideration. That being said, it is more interesting to look at the correct prediction of values of 1 (sensitivity), which appreciably changed from 0% in model A to 81.8% in model B. Fur-ther, specificity is 47.2%, positive predictive value 66,2% and negative predictive value 86.6%. From the results we can see that parking (p = .023) is shown to be statistically signifi-cant (p < .05) to the likelihood of customers shopping at Lidl. The odds ratio indicates for parking that an increase in one unit increases the odds of shopping at Lidl with 1.30. This means that an increasing importance of parking of the respondent is associated with an in-crease in the likelihood of shopping at Lidl for NBs. Under p < .1, atmosphere (p = .077) is statistically significant as well. However, coefficient B is negative (-.434), which indicates that the increase in one unit of atmosphere decreases the odds of shopping at Lidl with .648. This means that an increasing importance of in-store atmosphere to the respondent is associ-ated with a decrease in the likelihood of shopping at Lidl for NBs.

TABLE 2

Results binomial logistic regression analysis

Model A (N = 420) Model B (N = 106)

95% CI for odds ratio

95% CI for odds ratio

B SE Wald df p Odds Lower Upper B SE Wald df p Odds Lower Upper

28

Multiple regression analysis. The second part of the two-part model is a multiple

re-gression analysis of the six factors on SOW at Lidl. The models allows for a determination of the overall model fit and of the relative variable contribution to the total variance explained. The sample (N = 59) consists of all customers who currently spend money on NBs at Lidl (N = 53) and all respondents who currently shop at Lidl, but who do not spend any money on NBs (N = 6). These latter respondents are referred to as shoppers and most likely only buy PLs at Lidl. This sample increase might be beneficial to the reliability of the outcomes.

Eight assumptions are considered in order to run the multiple regression, with the first two considering the dataset. The model includes a continuous DV and six IVs measured at a continuous level. The other assumptions relate to how the data fits the model. First, there is no reason for the observations to be dependent. Second, the outcome variable is transformed us-ing ln(SOW/(1-SOW) in order to linearize the relationship of the dependent variable and to improve the model fit (Greene, 2003). Goal of this transformation is to enable predictors be-tween 0 and 1, which is necessary for a share variable. A visual inspection of a plot of studen-tized residuals versus unstandardized predicted values shows homoscedasticity. There is no reason to assume multicollinearity among the IVs as none of the variables have correlations > .5, and both the tolerance factor (>.2) and the VIF (< 5) show good results. There are no high-ly influential points (values for Cook’s distance >1). Moderate leverage points exist due to the weighted cases in the sample. It is also important to acknowledge that the values of the shop-pers may be considered as outliers by SPSS (studentized deleted residuals > +/- 3 standard deviations). Yet, it is decided to keep them in the analysis as these outliers are expected to benefit the reliability of our results. A histogram, P-P plot and a Q-Q plot demonstrate an ap-proximately normal distribution among the residuals. The only remark is that the sample could be considered small in comparison to the number of variables in the model. The guide-line of a minimum ratio of observations to IVs (5:1) is just met (9:1). Yet, this limitation is acknowledged and will be solved for in the subsequent analysis, which is discussed in section 4.3.2. Altogether, the necessary assumptions are satisfied.

The regression equation is expressed as follows:

Y(Share of Wallet Lidli) = Bo + B1 X1i + B2 X2i + β X3i + B4 X4i + B5 X5i+ B6 X6i + εi (4.3)

Where Y is the DV, B0 is the intercept, Bn are the slope coefficients, ε is the error term,

and “i” refers to the ith respondent of the 59 respondents. Xn are the IVs (X1 = Atmosphere, X2

= Duration, X3 = Assortment, X4 = Price/quality, X5 = Parking, X6 = Accessibility).