WILLINGNESS TO PAY FOR PRIVATE LABELS AND NATIONAL BRANDS AT HARD DISCOUNT RETAILERS

(1)

WILLINGNESS TO PAY FOR PRIVATE LABELS AND NATIONAL

BRANDS AT HARD DISCOUNT RETAILERS

A study on the effects of including premium private labels and national brands in a

hard discount retailer’s assortment

(2)

1

WILLINGNESS TO PAY FOR PRIVATE LABELS AND NATIONAL

BRANDS AT HARD DISCOUNT RETAILERS

A study on the effects of including premium private labels and national brands in a

hard discount retailer’s assortment

Master thesis

Msc Marketing Management & Marketing Intelligence

Bram van Santbrink Student number: 2201992 Nieuwe Boteringestraat 45D, 9712PH

Groningen

E-mail adress: bramvansantbrink@gmail.com Telephone number: +31634463471

Supervisor: J.E.M. van Nierop Secondary supervisor: J. van Doorn

University of Groningen

(3)

Management summary

As a result of the highly competitive retail market, hard-discount retailers try to differentiate themselves from their competitors by adding relatively expensive national brands to their private label dominated assortments. This thesis examines the benefits of the addition of these national brands. These benefits are compared to potential benefits that can be acquired by including premium private label products in a hard-discount retailer’s assortment.

By gathering a sample (n=1373) by means of an online questionnaire, this paper analyses the willingness-to-pay for private labels and national brands at hard discount retailers and their effect on willingness to pay for other items in their respective product categories. A representative sample of the Dutch population is collected by adding weights for each observation. By using information on product type and assortment varieties, models to predict the willingness to pay for 4 separate product categories were created. A latent class analysis was conducted to determine different effects of product type and assortment situation within the population. This analysis did not succeed in successfully segmenting the sample into distinguishable classes, but did provide model estimations for either 3 or 4 classes in each product category.

Consumers are found to be willing to pay significantly more for either leading and non-leading national brands in comparison to premium private labels. In similar fashion as non-leading national brands, premium private label availability may stimulate willingness to pay for other items within the same category. This effect is especially prevalent in 3-item assortments and in food categories. This category stimulating effect, as induced by premium private label availability, can exceed category effects that are caused by national brand availability. Also, 2-item assortments are generally better at maximizing the price consumers are willing to pay individual products’ than categories including more than 2 item choices. These findings may help grocery retailers in optimizing their assortment for each individual product category. It considers eliminating national brands or replacing them with premium private labels to maximize consumers’ willingness-to-pay and assortment attractiveness.

(4)

3

Preface

The completion of this thesis means more to me than acquiring the desired diploma. It is also the end of a large part of my life in which I have learnt more than I can imagine. Although I would not have envisioned myself to be where I am right now when I first started my studies at the university of Groningen, I am very grateful for all the experiences I have had and interesting people I have met.

(5)

1. Introduction

Over the past decade, the hard-discount (HD) retailing sector has captured a large percentage of the grocery shopping market. HD retailers like Aldi and Lidl are estimated to have grown their market share by up to 10% in a time span of only 5 years (Deleersnyder and Koll, 2012). The growth of these German retailers had motivated them to expand across borders. Due to their enormous success, they have recently been included in the top six retailers in western Europe (Planet Retail 2014). Their store concept consists of selling high volumes of a small range of relatively low priced and perceived lower quality products and offering little to no service in their stores (Wortmann, 2004). The

emergence of this HD format has affected the market share of the traditional retailer (TR). The HD retailer gains customers by targeting the price sensitive consumer. The large number of retailers that occupy the current grocery market has caused an inter-format competition in which both TR and HD stores compete for the same customer. Although this competition is not as fierce as the intra-format competition due to use of different pricing strategies (Cleeren et al., 2010), HD retailers seem to have initiated a battle for a larger market share. By offering lower prices than their TR competitors, they try to persuade the customers to switch from their current TR store to a nearby HD store. The availability of low priced products at the HD retailer therefore puts pressure on the prices of products in TR stores. This stresses cost savings in operational efficiency for TR stores to be able to stay competitive against the HD retailers’ threat (Van Heerde et al., 2008).

In the past, TR companies have had many advantages over HD retailers which caused them to target different customer segments. TR stores offer a deeper and broader assortment with many more stock keeping units (SKU) than the HD retailers. The two largest retailers in the HD category only carry an average of 1.770 (Lidl) and 1.440 (Aldi) SKUs compared to an average of more than 10.000 SKUs for TR stores (Nielsen, 2016). To be able to offer products at highly competitive price, HD retailers use the low stock keeping costs and low employee costs in comparison with their competitors. Also, because of their limited assortment they can optimize the benefits that are associated with economy of scale discounts (Ailawadi et al., 2010). The private label (PL) products that dominate HD retailers’ assortments contribute to the relatively low prices.

(8)

7 products, especially for price sensitive customers (Rubio and Yagüe, 2009), which may make it a very effective price tactic for HD retailers.

The assortment expansion does, however, not only come with potential benefits. The strategy holds risks for both the HD retailer and NB manufacturer. The manufacturer risks spoiling their relationship with other retailers. TR stores who sell the same NB can threaten to delist the NB from their store if the NB continues to be sold on shelves at the HD competitor (PlanetRetail, 2006). From the manufacturers’ perspective, more purchases of the NB product at a lower margin may not outweigh the larger margin that is gained from selling at the TR store (Deleersnyder and Koll, 2012). The HD retailer is also not without risk. NB sales may cannibalize their own PL sale and thus decrease profits (Deleersnyder and Koll, 2012). The prices of the newly introduced NB products can increase the price image of the store. This will cause consumers to perceive the HD retailer to be more expensive than they really are. The biggest problem caused by an inflated price image is that it may prevent price-sensitive consumers from visiting and buying at the store (Niedrich et al., 2001). Finally, HD retailers will be less independent as they will also have to deal with NB manufacturers’ wishes after including them in their assortment (Bergès, 2006).

Whereas many TR stores have started to combat the manufacturers’ dominance by introducing a three tiered private label offering to their assortment (Geyskens et al., 2010), HD retailers have yet to expand their PL-dominated assortment with their own products for multiple segments. The three tiered strategy involves introducing a PL option for three different segments to be able to serve each to their needs. This involves an economy PL product with the lowest quality for the most price sensitive customers, a basic PL product for consumers who are not willing to

compromise on quality for little extra savings and a premium private label (HQPL) product that offers the highest quality of all PL products in its respective product category. This HQPL item is more costly, but its quality should be at least on par with NB products.

The amount of HD stores is rapidly increasing. This results in an eagerness among

discounters to differentiate themselves from other HD stores (Deleersnyder et al., 2007). As variety within a brand line increases assortment appeal (Boatwright & Nunes, 2001), adding products of brands already available at the store may also be a viable option to attract new customers. Instead of adding more NB products to the assortment as multiple HD retailers are currently doing, the addition of high quality PL products may be a viable option to attract new upscale customers and increase the assortment attractiveness while remaining independent from brand manufacturers. The tiered approach will not be examined with three tiers as proposed by Geyskens et al. (2010), but will only research the availability of an economy and HQPL. This suits the nature of HD retailers in offering less SKU’s and brands than TR stores (Lourenço and Gijsbrechts, 2013), while also preventing any sort of choice overload (Iyenga and Lepper, 2000). This thesis aims to analyse the effects that NB products have on the HD retailers’ assortment from a consumer’s perspective and whether these benefits can be replicated by offering a HQPL product instead. This leads us to the following research question:

(9)

This study aims to contribute to the current area of research by examining the effects of HQPL offerings at HD retailers; a phenomenon that occurred in the Dutch retail environment fairly recently. Earlier research has mainly focussed on retailers’ PL offering for service retailers (Ter Braak et al., 2013; Ailawadi et al., 2008; Geyskens et al., 2010) , grocery retail in general (Pepe et al., 2012) or investigated consumers instead of the store format (Sethuraman & Cole, 1999; Bergès, 2006). So far there is no other article on the comparison between HQPL products and NB products at the HD retailer. The managerial implications are also huge. If the HD retailer will be able to expand its assortment attractiveness without third-party influences to comply with, they may become more independent, obtain larger profit margins and obtain more advantages in comparison to their competitors.

The next section will provide an overview of relevant research on this area of retailing, which leads to the proposal of a conceptual model. This is followed by a description of the research

methods describing the survey, participants and data collection. The penultimate section describes the results and finally the implications of this thesis is described in both research findings and managerial implications.

2. Conceptual background

This chapter discusses earlier research on the topic. Several subjects regarding retailing are discussed based on previous research. These subjects are believed to contribute to a deeper understanding of the HD retail environment and PL products. The hypotheses of this thesis are presented alongside with the literature review. This chapter concludes with a graphical presentation of the conceptual model.

2.1 Price and formats in the retail environment

Most grocery retailers in the Netherlands follow one of two retail formats. High-low (HiLo) strategies are preferred by TR stores. This format offers relatively highly priced products while running

frequent promotional campaigns (Wortmann, 2004). The stores offer a deep assortment which nowadays consist of many NB’s and PL products. NB’s tend to profit from stores using the HiLo format (Wortmann, 2004). However, following this strategy tends to hurt the PL shares across HiLo stores. For more PL-focussed stores like the HD stores, the every-day-low-pricing (EDLP) format seems like a more suitable choice. They offer products, as the name implies, at a constant relatively low price while keeping promotional frequency to an absolute minimum. This allows for a more constant product demand and reduces stock keeping costs to a minimum. The EDLP stores target consumers who are unwilling to travel to different stores to buy the best deals. While HiLo shoppers try to find the best deal at different stores, EDLP shoppers prefer to do all grocery shopping at the same store. This reluctance to visit other stores leads to larger average basket sizes at EDLP stores (Pechtl, 2004).

(10)

9 TR stores (Cleeren et al., 2010). No matter how small the current quality gap is across PL and NB products, Kumar & Steenkamp (2007) found that NB’s will often hold an advantage over PL products regarding perceived quality. This advantage is only slightly influenced by the actual quality of the products. Due to the direct link between the perceived quality of a PL brand to the retailer’s image (Olbrich et al., 2016), the retailer seeks to provide its customers with PL quality that is as high as possible to reduce negative carryover effects. The main reason consumers ascribe a higher quality to NB products is due to the NB’s reputation.

The different retail formats do not only affect the offerings in the stores, but also result in differences in customer preferences and characteristics across formats. EDLP shoppers are more price sensitive. They value that all prices are low at a single store and usually do all their grocery shopping at a single store. This results in larger shopping baskets due to lower store prices at EDLP stores (Pechtl, 2004; Mägi and Julander, 2005). HiLo customers are not focused on low pricing, but instead find quality and promotions to be of a larger importance. Both quality and the ability to go cherry picking among the best promotions are two of the most important drivers for HiLo shoppers’ behaviour. Olbrich et al. (2016) studied both shopper segments and found HiLo shoppers to be characterized by a higher risk averseness and the EDLP segment to be more price sensitive, which is in line with research by Mägi and Julander (2005). HiLo shoppers are therefore more inclined to buy NB products solely based on their reputation. In doing so, they avoid taking a risk on lesser known PL products. The opposite applies to the EDLP shopper segment. They are less risk aversive which, combined with the relatively high NB prices, attracts them to the PL products. They are however less store and brand loyal due to their high price sensitivity.

2.2 Price image in relation to products’ perceived costs

Actual prices and promotional activities are not the only factors influencing the consumers’

perception of the store and store choice. Nielsen (2008) reported that 70% of respondents to one of their surveys found it important that EDLP stores have a reputation for being cheaper than others, even if this is not the case regarding their actual prices. This only further strengthens the belief that what is perceived by the customer may well be different than the message the retailer is trying to convey with its actions. Baker et al. (2002) identified three different types of environmental cues that influence store and assortment value perceptions:

 Social cues refer to the perceptions the consumers form of the store employee and whether the store is crowded or not. Unfriendly employees or a stressful crowded environment may cause the consumer to incur shopping experience costs. These influence consumers’ perceptions on the store’s assortment value. A drop in the assortment value may cause the price image to rise; consumers feel like they have to pay too much money for the service the store provides.

 The ambient factors refer to the store environment in total. If facilitating cues, like music or store smell, are perceived as favourable they can reduce perceived costs like time spent in the store. This can also indirectly lead to an increase in perceived assortment value.  Finally, the design refers to store design and lay-out. Canter (1983) describes that for many

(11)

lead to an increase of the price image or reduce the perceived assortment value (Baker et al., 2002).

Hamilton & Chernev (2013) support the notion that consumers rely mostly on in-store cues to form an opinion on whether prices are high or low. This is in line with earlier findings that the within-store price gap between products has a larger role in this image formation than the between-store price gap. Consumers form such a mental image based on how expensive a between-store is in their experience (Lourenço et al., 2015). By doing so, they rely on both price and non-price cues and are influenced by earlier experience. Informed consumers will adjust earlier set reference prices to this price image and evaluate prices according to the adjusted reference price. Uninformed customers will however base their price evaluations solely on the price image, as they have not yet have the experience to form a reference price (Hamilton & Chernev, 2013). This is an important distinction as uninformed customers will be more easily inclined to conclude that all prices at low-price-image stores, like HD stores, are low. As consumers unfamiliar with HD retail stores are more easily persuaded into adjusting their price image of the store, respondents in this study will be asked for past and current shopping experiences at HD retailers to see whether familiarity with HD stores has any influence on regarding price estimations of HD retailer products.

2.3 Different effects of private labels across product categories on

price image formation

Almost a decade ago, PL products were already available in over 90% of all product categories and accounted for a significant market share in Europe: 23% (TNS, 2009). This market share was already much larger in some western European countries such as the United Kingdom, Switzerland and Germany who respectively showed a 46%, 45% and 37% PL market share (Europanel, 2009). Although this value is slightly lower in the United States, a large part of all grocery retail sales in the Western world already involved PL products (PLMA Yearbook, 2007). PL product penetration and their value market share do show differences between product categories. Hoch and Banerji (1993) investigated PL market shares across categories. They identified several factors that help in

predicting whether PL brands are able to obtain a large market share. PL brands did best in

categories that generate a large amount of sales, where few NB products are situated and where the NB manufacturers that are present have small advertising expenditures. Although the last two indicators are likely to facilitate PL penetration and value market share success by directly influencing the competing products, the first indicator may be related to the fact that PL brands greatly benefit from economies of scale to provide consumers with a competitive price. Bergès et al. (2012) adds that consumer trust could also have an important influence on PL penetration and success. Although PL quality is usually not much worse than their NB counterparts (Cleeren et al., 2010), which will be discussed in the following section, the good reputation of the NB alternative will more easily persuade consumers in categories where consumer trust is essential, like baby nutrition.

(12)

11 expensive categories do exert a larger influence on the formation of the price image. The often larger price gap in expensive categories may cause the consumer to perceive the stakes to be higher to make the right decision. Decreasing prices on some of these categories while trying to retain a similar level of quality may therefore positively influence the price image of the HD retailer and seem like ideal categories to introduce further PL products in.

Building on earlier studies by Hamilton & Chernev (2013), Lourenço et al. (2015) analysed consumer purchase behaviour and found that consumers were more influenced by prices of products that could be bought in large quantities (i.e. for storing and later usage) and for expensive items. Although Hamilton & Chernev (2013), and earlier Desai and Talukdar (2003), did find a positive effect of purchase frequency on the importance of a category on price image formation, Lourenço et al. (2015) was unable to confirm this result as he found a small negative relation between purchase frequency and price image formation at TR stores (β=-0.014). The negative relation is explained by providing a behavioural explanation: frequently bought products are supposedly bought habitually. A price change would therefore have little influence on the price image of the store. The fact that inexpensive food products are bought habitually has later been supported by Olbrich et al. (2016). It seems that PL success as measured by penetration and market share is highly variable across product categories. This can imply that some product categories are less suitable for economy, basic and HQPL products.

2.4 Perceived differences between NB and PL and their marketing

strategies

NB products are made by manufacturers who distribute the product nationally across stores. Manufacturers try to combat the increasing market share of PL products by innovating their products and by using large marketing budgets to increase sales (Deleersnyder, 2007). Although including NB products is crucial in relation to assortment attractiveness and profitability, as

consumers usually prefer NB over PL products (Ailawadi, 2001), retailers benefit more from sales of their own PL brands. This can be explained due to the fact that PL product provide a larger profit share comparison to NB products in the same categories (Sudhir & Talukdar, 2004). The strong NB’s presence are also very important in maximizing PL sales. NB’s attract NB-loyal customers to the retailer (Pepe et al., 2011) and removal of such products from the assortment may decrease store profitability.

(13)

probability of purchasing the basic PL at retailers that have not yet implemented the three tiered PL strategy (Dhar and Hoch, 1997). However, the effect of demographics on economy and premium products’ purchase intentions has not yet been researched.

Although many HD discounters are yet to fully capitalize on the segmented PL strategy, implementing this tactic was already a hot topic among TRs several years ago (Kumar & Steenkamp, 2007). The introduction of other PL product types can have large influences on present PL products at the HD retailer, but it is also interesting to investigate the effect on the recently introduced NB assortment at the HD retailer. Due to contradicting findings regarding demographics’ influence on basic PL sales (Dhar and Hoch, 1997; Bergès et al., 2009), sales drivers may also vary across the other PL product tiers and store formats.

2.5 Offering national brands at hard-discount retailers

A current trend for HD retailers is expanding their PL dominated assortment with NB products. This allows the discounter to continue its growth by retrieving its customers from multiple segments (Deleersnyder et al., 2007). These HD stores used to be divided into two different classes (Cleeren et al., 2007). In the past, consumers could distinguish between hard- and soft limited line discounters (Aggarwal, 2003).The hard limited-line discounters, like Aldi, used to sell almost exclusively PL items while the soft limited-line discounters like Lidl and Netto included only a limited set of leading NB’s. Although both segments have started adding leading NB’s to their assortment blurring the current distinctions between both segments, some discounters have also added many non-leading NB products to their assortment. Despite non-leading NB’s profiting from the superior quality bias in comparison to PL products (Olbrich and Jansen, 2014), there is a strong difference in assortment perceptions and store performance between leading and non-leading NB introductions. The quality bias is induced by the expensiveness of the NB compared to the PL (Olbrich et al., 2016). This price-quality bias causes consumers to implicitly ascribe a higher price-quality to more expensive products. Non-leading NB introductions often fail to produce a positive category or store effect. This is in contrast with leading NB’s. These often do generate favourable effects for the retailer (Lourenço and Gijsbrechts, 2013). By assessing the effect NB introductions have on a HD retailers assortment in both price image and assortment perception, Lourenço and Gijsbrechts (2013) analysed over 4000 households who shopped at either Lidl or Aldi in Belgium. By investigating store image data and share-of-wallet expenses at the HD retailers, they discovered that leading NB’s can charge prices up to 2,5 times as high as PL without causing negative effects on the price image of the retailer, but non-leading NB’s will have negative effects when charging more than 1,7 times the price of the available PL product. Value and assortment perceptions are therefore also found to be heavily dependent on product leadership of the NB.

Earlier research by Deleersnyder et al. (2007) shone a light on the many positive effects of adding NB products to the assortment of the HD retailer. Not only does assortment attractiveness and perceived assortment quality and value increase, but NB loyal customers are also attracted to the store. Whether a consumer’s favourite brand is available at the store is also an important factor affecting consumer’s perception of the assortment (Broniarczyk et al., 1998). Although usual PL buyers may switch to purchasing NB products and cause profits to decrease, Deleersnyder et al. (2007) found most NB introductions turn out to be beneficial to both the discounter and

(14)

13 the manufacturers’ best interest to enter the HD retailer’s assortment with a modest NB with a lot of potential to limit store switching consumers who do their usual grocery shopping at a TR store. Results of Olbrich et al. (2016) show this is not in the retailers’ best interests. The lesser performance of non-leading NB’s in comparison to leading NB products may hurt store performance. As there is only little information available on the differences between leading and non-leading NB product influences at HD retailers, the following is expected based on aforementioned differences within NB products research:

Hypothesis 1: Leading national brand presence will increase willingness-to-pay of additional economy private label and non-leading national brand products in the assortment. Hypothesis 2: Consumers’ willingness-to-pay for leading national brands is significantly higher than willingness-to-pay for non-leading national brands.

Pricing of NB products at HD retailers is a controversial subject. The between-store price gap of NB products only has a relatively small effect on a retailer’s price image in comparison to the in-store price gap of NB and PL products. Many HD retailers therefore only gain small benefits from consumers purchasing the NB, but benefit most from having a large gap between the NB and their PL products (Deleersnyder et al., 2007). A large within-store price gap increases demand for the lower priced PL product. This synergizes well with the higher price sensitivity that is displayed by the consumers who shop at EDLP stores, like HD retailers (Rubio and Yagüe, 2009). Other research has also supported the finding that enlarging the price gap between NB and PL products increases PL market shares (Sethuraman and Cole, 1999). Although the willingness-to-pay for NB’s remains higher, the large price gap is incentive for consumers to switch to the PL. Ailawadi et al. (2010) state that only large retailers are able to build such a price gap by exploiting their economy of scales advantages to reduce the PL prices. The low number of unique SKUs in the HD stores could facilitate these economy of scale advantages (Wortmann, 2004).

(15)

following:

Hypothesis 3: Economy private label availability in non-food categories increase willingness-to-pay for higher quality products.

The downside of having large price differences is due to the effect that highly priced products have on the store price image. This is where an opportunity may present itself for HD retailers. While adding a highly priced NB to the assortment to boost its assortment value perception and expand its customer base into less price sensitive segments may have some positive effects, why not add a HQPL product of similar quality?

2.6 High quality private label products as a solution?

The economy PL is already present in the current HD retailers’ assortment and it has been for a long time. The standard PL offering at HD retailers is regarded as economy PL products, because of their bottom-of-the-line pricing and not having a NB counterpart with similar quality (Burt, 2000); their quality is acceptable, but not great due to the products having to be offered at the best price as possible (Geyskens et al., 2010). HQPL products are of higher quality than the normal PL products that are already offered, but they also come at a price premium. This price premium can be much smaller than what NB products are offered for, which decreases the risk of blurring the discount positioning of the price image of the HD retailer (Deleersnyder et al., 2007). This reduces one of the dangers HD retailers risk for adding NB products to their generally lowly priced assortments. The increased price of HQPL items can however still be used to retain a large in-store price gap to signal the especially low price of the economy PL (Olbrich et al., 2016). This could retain the positive aspects of having highly priced products, as the retailer might still be able to attract other consumer segments while reducing the risk of blurring the intended price positioning.

Despite the possible advantage NB products have over PL products in terms of perceived quality (Kumar and Steenkamp, 2007), PL lines have constantly been improved by gaining more control over the production process and the associated costs. This has resulted in a very small actual quality gap between PL and NB products that is magnified by both a price-quality bias and the general belief that NB products are of higher quality (Olbrich and Jansen, 2014; Kumar and

Steenkamp, 2007). As the purchase risks as experienced by consumers increase, the preference for NB products, thus the willingness-to-pay for these products, will most likely increase. Based on category differences between food and non-food products and expensiveness and the price-quality bias (Olbrich et al., 2016), the following hypotheses are created:

Hypothesis 4: Non-food leading national brands are associated with a higher willingness-to-pay than non-food high quality private label products.

Hypothesis 5: Non-food non-leading national brands are associated with a higher willingness-to- pay than high quality private labels.

Hypothesis 6: Within the food category, Willingness to pay for non-leading national brands will not differ from willingness-to-pay for high quality private labels.

(16)

15 Adding HQPL products also has other benefits for the retailer. In contrast to NB products, HQPL products provide the retailer with a similar healthy profit margin as their other tiered

counterparts. Bonfrer and Chintagunta (2004) found standard PL products to have a profit margin of 20% higher on average than the NB assortment in a mixed sample of EDLP and HiLo stores over a period of 2 years. Having a high profit margin is already an important factor for retailers to expand their PL assortment (Ailawadi, et al., 2008), which further strengthens the idea of introducing HQPL instead of NB products. In store brand cannibalization will be less of a problem due to all available brands providing a good profit margin for the retailer. As many private labels are already exclusively available at HD retailers, the HQPL introduction will only further add to the uniqueness of the assortment. Exclusiveness was also found to increase store loyalty in earlier research by Corstjens & Lal (2000). The uniqueness and exclusivity of offering HQPL products leads us to develop the final hypothesis:

Hypothesis 8: High quality private label availability will increase within category willingness-to-pay in comparison to assortments without high quality private labels.

2.7 Conceptual model

All eight hypotheses are visualized in the conceptual model below in figure 1. This conceptual model displays all hypothesized relationships and constructs. This research will test whether the

independent variables, national brand availability, economy private label availability, high quality private label availability and the differences across product categories and product leadership will have an effect on the willingness to pay of the sample. These measurements are controlled for favourite brand availability. There are some aspects of the framework that need further explanation to be fully comprehended. These will be described in the next chapter during the operationalization of the elements of the conceptual model.

Figure 1: Conceptual model with all hypothesized relationships

3. Methodology

(17)

description for several constructs that have not been described sufficiently will be provided.

Subsequently, the data, survey, participants, research method and data cleaning will be discussed to enable to possibility of replicating this study.

3.1 Operationalization of constructs

This section will briefly describe several constructs that are displayed in figure 1 on the previous page. These aspect need further explanation to be fully comprehended and incorporated into the conceptual model.

3.1.1 Willingness-to-pay

Willingness-to-pay (WTP) reflects the maximum price a consumer is prepared to pay for a product (Wertenbroch and Skiera, 2002). It is assumed that WTP is directly linked to the attractiveness of a product. An increase in eagerness to obtain a product can be observed as an increase in WTP for that product. The WTP measures the value a consumer ascribes to the product. In line with this reasoning, a consumer’s WTP is a representation of a product’s subjective attractiveness. As the research was not conducted in a real supermarket, no purchase data to deduct the real WTP from was obtained. Miller et al. (2011) has tried measuring WTP in 4 different ways and compared these measurements to the real WTP that was acquired through an online store. They found all sources of WTP to be biased in comparison to the consumers’ real WTP. Incentive aligned options provided estimates of WTP that were closer to the real WTP than the hypothetical methods. As hypothetical methods are found to predict WTP ratings fairly well despite their bias and due to the absence of coercive means as described in Miller et al. (2011), this study aims to measure hypothetical WTP through an open ended format. This research aims to measure whether the WTP for a HQPL is significantly different from a NB product, but also aims to examine whether adding a HQPL product to the assortment has any effect on WTP ratings of products other than the HQPL. This will be tested through hypothesis 8.

3.1.2 Product category

As many previous research studies have found a variety in PL attractiveness, penetration and success across categories (Briesch et al., 1997; Narasimhan & Wilcox, 1998; Hamilton & Chernev, 2013), it is important to account for category effects as best as possible. Lourenço et al. (2015) identified several very influential categories at HD retailers to determine the price image of a store. Hamilton & Chernev (2013) found that often purchased categories matter more to consumers as they are more often confronted with these prices. Moreover, more often purchased categories have a larger influence in altering the experienced shopping utility (Bell, et al., 1998; Inman, et al., 2009). Due to these across-category differences, this study should assess the effects of different categories when comparing the relationship between WTP for both NB and PL products.

3.1.3 Product leadership

(18)

17 in the product categories should therefore also include both leading and non-leading NB’s to

preclude any brand specific bias. 3.1.4 Favourite brand

Whether a respondent named one of the brands used in the experiment as his/her favourite brand is also likely to influence WTP ratings. Favourite brand availability influences store choice probability (Briesch et al., 2009) and may also contribute to the overall attractiveness of the store’s assortment (Broniarczyk et al., 1998). It is therefore expected that respondents who find their favourite brand in the experiment are likely to provide higher WTP ratings for these products than consumers who are indifferent. The analyses will therefore be controlled for favourite brand availability.

3.2 Data collection

Research participants where gathered by distributing an online questionnaire. The questionnaire was pre-tested by several students before it was distributed among the population. An internet link to the survey was distributed through social media and shared on different internet forums. This internet link was valid for 11 days, in which all respondents were able to complete the survey. Distributing the link on these forums was found to be much more effective than gathering participants through the researcher’s personal network. For those who were participating in the study, there was a possibility of being awarded one out of two €10 gift vouchers. Only those who fully completed the survey had a chance of winning a gift coupon. The purpose of this voucher was to motivate more people to fill in the survey and the chance of winning was independent of the answers that were given in the survey. The two winners were picked at random.

3.3 Online questionnaire

This part will describe the different components of the online survey. Both Dutch and English versions were created to ensure that foreigners living in the Netherlands were also able to fill in the survey. The default language used in the survey was Dutch. All participants were told that their responses would be handled completely anonymous and that the survey would take about 5 minutes to complete. The average time spent on completing the survey was 6 minutes and 42.4 seconds and a median response time of 6 minutes and 3 seconds across 1323 respondents. The English version of the survey can be found in appendix 7.1.

3.3.1 Shopping and demographic characteristics

Participants were asked to answer a few questions regarding their shopping behaviour. This includes questions concerning the number of times respondents go shopping for groceries every week, their favourite store and where they usually go shopping for groceries.. Furthermore, participants were required to answer questions regarding their demographic situation at the end of the questionnaire. This covers basic demographic variables, like age, gender, marital status, employment and

education. These variables might later on be used to segment the research sample and ensure the external validity of the results.

3.3.2 Willingness-to-pay for grocery items

(19)

which they were asked to state their WTP. This does not concern their real WTP, but their

hypothetical WTP following earlier research by Miller et al. (2011). Respondents were told to take into account that these were the only available items at the Lidl they were doing their grocery shopping. The situation sketch respondents were presented with states the following:

“Lidl currently has its own luxury brand: Lidl Delicieux and Lidl Deluxe. In the past, this brand was only available during the holidays, but it is currently available all year round. The Lidl

Delicieux/Deluxe brands are now offered simultaneously with the usual Lidl brands. The next questions will each show a Lidl assortment. You will be asked how much euros you are willing to pay for each product in the assortment. Imagine that the products that are shown are the only available products in this product category. There are no available alternatives.”

Participants were shown 2 assortments for each of 4 products categories. This amounts to a total of 8 sets of products that were shown to each participant. Several rules were used to

determine which product categories would be used for the experiment. The product categories should already contain one leading NB and one or more non-leading NB’s in the Dutch grocery retailing environment. This rule was set to provide participants with a situation that is as realistic as possible. To determine if there is an effect of product categories and product leadership as depicted by hypotheses 4 up until and including 7, the product categories were divided in 2 food and 2 non-food categories. These product categories should also be available across retailers including Lidl to maximize external validity. The choice was made to use two product categories for both food and non-food products:

 Pasta sauce & chocolate sprinkles  Liquid laundry detergent & shampoo

The NB products that were included in the product sets were chosen on account of the brand ranking by Foodmagazine and Distrifood (2016). The product categories and the corresponding brands, some with corresponding ranking in the retail brand top 100, can be found in table 1 below. The questionnaire was set up in such a way that every respondent was shown 2 product combinations of each product category. The number of available brands in this product category was randomized for either 2, 3 or 4 available brands. There was no possibility in which a participant could be presented with a set of the same products as they were only presented with 2, 3 or 4 choice options once per product category. For example, if a participant was presented with 2 choices of pasta sauce in the first product assortment, the next product assortment of pasta sauces could only contain 3 or 4 choice options.

Product category

Pasta sauce Liquid laundry detergent Chocolate sprinkles Shower gel Leading national brand Grand’ Italia (55)

Robijn (34) De Ruijter (87) Andrélon (100)

Non-leading national brand

Bertolli (94) Omo Venz Fa

High quality private label

Lidl Delicieux Lidl Delicieux Lidl Delicieux Lidl Delicieux

Economy private label

Lidl Italiamo Lidl Formil Lidl Mister Choc Lidl Cien

(20)

19

3.4 Participants

To ensure optimal validity of the gathered sample, the demographic variables were compared to the demographical information of the Dutch population that was gathered by the Central Bureau of Statistics (2016). A comparison of the sample and the Dutch population can be found in table 2. The skewed distribution of gender between the population and the research sample does not necessarily decrease the external validity of the findings. Many researches indicate the female dominance regarding shopping and associate women with the activity of shopping instead of linking the activity equally to both genders (Lunt and Livingstone, 1992; Bertrand and Davidovitsch, 2008). However, as the current sample sees female responses outweigh male responses by more than sixfold, a

correction will be made to increase weights of male responses.

Due to the composition of the researcher’s personal network and distribution by online means, the sample is underrepresented in the two most elderly age classes. This may have to do with elderly consumers not being on social media, not being in the researcher’s network and having inadequate experience with the internet to be able to fill in the online survey. Increasing the weights of the observations of participants aged 65 and older and decreasing the weights of participants of 40 and younger should give a more accurate representation of the Dutch consumer population. Using these weights will probably also resolve any generalizability issues regarding marital status as elderly are more common to be widowed than younger people are. The unweighted age distribution can be found in figure 2 on the next page.

Although there are no exact numbers of the total Dutch population regarding the measured levels of education that were used in this study, it is possible to compare the research sample with data of the central bureau of statistics. The participants are ascribed to one of three groups used by the central bureau of statistics (2016): Lowly educated, intermediately educated and highly

educated. Using their definitions provided a weighted and unweighted distribution of education levels of the research sample. Comparing the weighted distribution, while using the weights of both gender and age as discussed above, to the Dutch population, a slight overrepresentation of highly educated respondents and a slight underrepresentation of lowly educated respondents is

encountered. This difference is large enough to threaten the external validity and therefore warrants further data transformation.

Variable Variable level Dutch population (CBS, 2016) Research sample Gender Male 49.6% 13.2% Female 50.4% 86.8% Age >20 22.5% 8.3% 20 until 40 24.5% 51.2% 40 until 65 34.8% 37.3% 65 until 80 13.8% 3.0% 80 and older 4.4% <0.0% Marital status Unmarried 47.8% 55.7% Married 39.6% 36.1% Widowed 5.1% 1.9% Divorced 7.5% 6.3%

(21)

The final aspects for which the representativeness of the sample can be improved concerns the employment statuses. This can also be compared to data provided by the Central Bureau of Statistics (2016). For this assessment, the assumption is made that none of the student respondents is currently looking for a job or working more than 12 hours a week. These respondents are

therefore classified as not being part of the labour force of the country. An overrepresentation of unemployed respondents and an underrepresentation of full time employed participants is observed in the research sample. Therefore a slight correction is made regarding unemployed and full time employed participants. The final weighted responses for all demographic variables can be found in table 3. The used weights for this transformation can be found in table 4 on the next page.

Application of these weights to our dataset greatly improved the fit of the used sample to the Dutch population. As a consequence of this better fit, the external validity of our results will also be improved. For the weighted and unweighted samples, appendix 7.2 provides a weighed age distribution of the sample.

Dutch population* Unweighted research sample Δ Population-unweighted sample Weighted research sample Δ Population-weighted sample Improvement / deterioration due weighing Gender Male 49.6% 13.2% 36.4 19.2% 30.4 +6.0% Female 50.4% 86.8% 36.4 80.8% 30.4 +6.0% Age >20 22.5% 8.3% 14.2 10.5% 10 +4.2% 20 until 40 24.5% 51.2% 26.7 37.1% 13.6 +13.1% 40 until 65 34.8% 37.3% 2.5 37.2% 2.4 +0.1% 65 until 80 13.8% 3.0% 10.8 14.2% 0.4 +10.4% 80 and older 4.4% <0.0% 4.4 1.1% 3.3 +1.1%

Marital status Unmarried 47.8% 55.7% 7.9 48.8% 1.0 +6.9%

Married 39.6% 36.1% 3.5 39.8% 0.2 +3.3%

Widowed 5.1% 1.9% 3.2 5.1% 0.0 +3.2%

Divorced 7.5% 6.3% 1.2 6.3% 1.2 0

Employment Unemployed 4.2% 15.2% 11.0 10.2% 6.0 +5.0%

status Part time employed 31.9% 33.0% 1.1 28.3% 3.6 -2.5% Full time employed 33.9% 24.6% 9.3 25.0% 8.9 +0.4% Not part of the labour force 30.0% 24.2% 5.8 36.5% 6.5 -0.7% Education Low 30.4% 20.8% 9.6 30.7% 0.3 +9.3% level Intermediate 39.9% 43.1% 3.2 37.6% 2.3 +0.9% High 29.7% 36.1% 6.4 31.8% 2.1 +4.3% Total 193.6 122.6 +71.0%

Table 3: Overview of unweighted and weighted responses across demographic variables. *Demographic data acquired through the CBS (2016)

(22)

21

3.5 Data cleaning

As many answers to open questions were spelled incorrectly, these had to be corrected to enable analyses of the answers. Both abbreviations and misspelled answers that resembled a major grocery retailer in the Netherlands were adjusted to the right format.. A similar correction was made when respondents had to provide their favourite brand of pasta sauce, laundry detergent, chocolate sprinkles or shampoo brands. Many answers resembled either a leading NB, non-leading NB or PL that was used in the experiment, but was spelled incorrectly. These answers were therefore also corrected accordingly.

The data was subsequently checked for outliers. Responses that took longer than 1300 seconds were removed from the sample due to them taking too long to fill in the questionnaire. Similarly, responses that took less than 180 seconds were removed due to it being highly unlikely that these respondents filled in their answers accurately. Those who took less than 8 seconds to fill in their demographic information for the WTP task ( = 21.5, σ=13.9) or took longer than 70 seconds were also removed.

To avoid any statistical errors regarding the WTP-ratings, outliers in the dataset were removed in accordance with Tukey (1977), choosing to remove all observations with WTP ratings outside of the 1.5 interquartile range. Although this option may lead to a larger loss of variance, the dataset is sufficiently large enough to cope with the loss of several additional observations. In comparison, removing outliers based on the 3.0 interquartile range option leads to removing 114 additional outliers in comparison to Tukey (1977). To make sure respondents who are not interested in the specified product categories do not skew the distribution, the most extreme values possible (0.0 and 8.0) were also deleted from the observations. In total 986 outliers were removed out of which 310 fell in the pasta category, 191 for the laundry detergent category, 242 belong in the chocolate sprinkles category and 243 fit in the shampoo category. The large amount of outliers in the pasta category can partly ascribed to several comments of respondents who said that they never buy pasta sauce or make their own pasta sauce. This resulted in them giving a rating that expressed the respondent did not want to pay any amount for all shown brands of pasta sauce. Also, it is likely that the lesser amount of outliers that were removed in the detergent category can be ascribed to

Variables Weights Gender Male 1.5 Female 0.9 Age <20 0.7 20-40 0.5 40-65 0.7 65-80 3.0 80+ 8.0 Employment Unemployed 0.7 Full time employed 1.1

Education level Low 1.3

High 0.9

(23)

the larger standard deviation of WTP ratings in this category. The exact number of outliers that were removed for each brand can be found in appendix 7.3. The information pertaining which

combination of brands were shown for each product set is still registered in four newly created dummy-variables. This results in the creation one dummy variable for each product type that was shown with each given WTP rating:

 Leading national brand  Non-leading national brand  High quality private label  Economy private label

3.6 Plan of analysis

To test for the truthfulness of the 8 hypotheses, a model will be created and tested in both a multiple regression analysis and latent class analysis. Tools to be used for these analyses are Latent Gold and R-Studio. Before conducting the multiplicative regression analysis, the explanatory variables in the model should be tested for multicollinearity of the variables and heteroscedasticity and normality of the residuals . To test whether the residuals are normally distributed, both the Kolmogorov-Smirnov, as described by Lopes, Reid and Hobson (2007), Jarque-Bera test (Jarque & Bera, 1987) and Lilliefors test will be conducted. Test for heteroscedasticity will concern the Levene’s test and the multicollinearity will first be checked be analysing the GVIF scores of the variables.

After the residual analysis, the model should adhere to all earlier mentioned model

assumptions. If this is not the case during initial analyses, transformations will be done to the data to shape the data in the preferred shape. At first, the dataset contained of 1323 unique respondents. After removal of outliers and transformation of the data 29.529 responses were obtained and 1317 unique respondents were used in the analyses. Each observation contains the characteristics of the respondent, the characteristics of the product the respondent was rating, the item set in which the product was shown and, finally, the WTP rating that was given for this specific product.

Subsequently a regression model will be provided to test the hypotheses. After the multiple regression analysis, which uses ordinary least squares, a latent class analysis will be done to

distinguish different customer segments. Latent Gold will provide a latent class regression model to divide the sample into segments based on unobserved characteristics. The ideal number of latent classes will then be determined by analysing information criteria as provided by the latent class analysis.

4. Results

(24)

23

4.1 Model estimation

The hypotheses and measured parameters have provided us with a linear additive model. Firstly, the model is presented and various variables will be explained. Secondly, discuss the parameter

estimates and various p-values as provided by the model will be discussed. This will also include remarks on the model fit. The model takes the form of a summation of explanatory variables that predict the price consumer i is willing to pay for product j:

𝑃_𝑖𝑗∗ _{= β}

1𝐶𝑖𝑗+ β2 𝑇𝑖𝑗+ β3 𝐶𝑖𝑗𝑇𝑖𝑗+ β4𝐷1_𝑖𝑗+ β5𝐷2_𝑖𝑗+ β6𝐷3_𝑖𝑗+ β7𝐷4_𝑖𝑗+ 𝜀𝑖𝑗 (4.1) The equation components of model 4.1 are described in table 6 below. Due to the fact that this model does not allow us to make any interpretation regarding the different situations that the items were shown in, another adjusted model was created. This model does not include the 4 dummies, but uses a factorial variable depicting the assortment situation the item was shown in instead. This factorial variable, ‘Situation’, is made up of 11 different assortment situations. There are 6 possible 2-item assortments, 3 different 4-item assortments and 1 4-item assortment. The used variable coding can be found in appendix 7.4. Following this logic, leading NB availability is coded as NB+, non-leading NB availability as NB-_{, HQPL availability as PL}+_{and economy PL availability as PL}-_{. When}

both NB product types are available, their availability is written as NB2_{. This is similarly done for}

availability of both PL product types. This is described with the code PL2_{. In line with this reasoning, a}

3-item assortment consisting of leading NB, non-leading NB and economy PL would be coded as NB2_PL-_{. In similar fashion, a 2-item assortment of the non-leading NB and HQPL is described by using}

NB-_PL+_{as code. The formula of the model in which this situation variable is included is displayed}

below:

𝑃_𝑖𝑗∗ = β1𝐶𝑖𝑗+ β2 𝑇𝑖𝑗+ β3 𝐶𝑖𝑗𝑇𝑖𝑗+ β4𝑆𝑖𝑗+ 𝜀𝑖𝑗 (4.2) The contained equation components for model 4.2 can also be found in table 6 below.

Model 4.1 Model 4.2

Equation component

Description Equation

component 𝑷𝒊𝒋∗ A Box-Cox transformed willingness-to-pay

rating consumer for respondent i for product j

𝑃𝑖𝑗∗ A Box-Cox transformed willingness-to-pay rating

consumer for respondent i for product j 𝑪𝒊𝒋 Factor indicating the product category of

product j that the rating refers to. This factorial variable refers to the product categories (i.e. pasta sauce) that were analysed.

𝐶𝑖𝑗 Factor indicating the product category of product j

that the rating refers to. This factorial variable refers to the product categories (e.g. pasta sauce) that were analysed.

𝑻𝒊𝒋 Factor indicating the product type of product j

that the rating refers to. This factorial variable refers to the product types (i.e. leading national brand) that were analysed.

𝑇𝑖𝑗 Factor indicating the product type of product j that

the rating refers to. This factorial variable refers to the product types (i.e. leading national brand) that were analysed.

𝑫𝟏 Dummy indicator for leading national brand

presence in the shown product set.

𝑆𝑖𝑗 Factor describing the item set that was shown in

combination with product j for respondent i 𝑫𝟐 Dummy indicator for non-leading national

brand presence in the shown product set.

𝛽1− 𝛽4 Regression coefficients for their respective

explanatory variables 𝑫𝟑 Dummy indicator for high quality private label

presence in the shown product set.

𝜀𝑖𝑗 The disturbance term for predicting the response

variable for respondent i for product j 𝑫𝟒 Dummy indicator for economy private label

presence in the shown product set. 𝜷𝟏− 𝜷𝟕 Regression coefficients for their respective

explanatory variables

𝜺𝒊𝒋 The disturbance term for predicting the

(25)

4.2 Residual analysis

To infer if any deductions can be made from both models, there are several criteria the models should comply with. After removal of all outliers, the model should adhere to several important rules as depicted by Leeflang et al. (2015). To decrease the data bias to a minimum, the model was

checked for the following 5 assumptions:

1. The expected value of residuals for all observations should be equal to 0. 2. Normal distribution of error terms

3. No autocorrelation across observations 4. No multicollinearity of explanatory variables 5. Homoscedasticity across error terms

As autocorrelation is only present when there is a pattern in the residuals over time, autocorrelation is only possible if data is collected over time (Leeflang et al., 2015). As the data concerns a cross-sectional sample, autocorrelation will not be present and will therefore not be tested. The other 4 residual analyses will be discussed in this chapter.

4.2.1 Expected error value equals zero

The first assumption suggests that the predicted value of the error term in the models should equal 0. The RESET-test, as suggested by Ramsey (1969), can test whether the linear form of the models is correct or whether there is an effect unaccounted for. After testing the both model 4.1 and 4.2 with this test, the results showed an insignificant p-value for both model 4.1 (F =0.44, p = 0.64) and model 4.2 (F = 2.12, p = 0.12). This indicates that there are no issues with the functional form of the

models. This distribution of residuals indicates a model transformation is not required. 4.2.2 Normality

The second assumption to test is whether the residuals of the models are normally distributed. The residual distribution of both models can be found in appendix 7.5. Although a first visual inspection of the residuals does not provide indication of non-normally distributed residuals, it does show many residual abnormalities near both ends of the spectrum for both models. To test whether the

residuals are normally distributed the residuals should be tested with either the Kolmogorov-Smirnov test or the Shapiro-Wilk test. As the Shapiro-Wilk test is limited at 5000 observations in R, the Kolmogorov-Smirnov test is performed to test for normality of the residuals. The null hypothesis of this test states that the residuals are normally distributed. Regarding model 4.1 the Kolmogorov-Smirnov test turned out to be significant (χ2 = 0.039, p < 0.01), indicating the residuals are not normally distributed. A similar observation was provided for model 4.2 (χ2 _{= 0.057, p < 0.01).}

(26)

25

𝑃

_𝑖𝑗∗

=

𝑃𝑖𝑗

λ₋₁

λ (4.3)

After applying a Box-Cox transformation the Kolmogorov-Smirnov test still indicated non-normality (χ2_{= 0.057, p < 0.01), just like the Jarque-Bera (p < 0.01) and the Lilliefors test (p < 0.01)}

for model 4.1. Model 4.2 followed a similar result as it also did not improve after the transformation. All three tests statistics, the Kolmogorov-Smirnov (χ2_{= 0.054, p < 0.01), Jarque-Bera ( p < 0.01) and}

Lillefors test (p < 0.01), remained insignificant. As both models are still influenced by non-normal residuals, a different option is explored to remove outliers in the dataset to stimulate normality. To improve the normality of the residuals, the further removal of outliers in the dataset was tested. As the dataset consists of ratings of all four product categories, the most expensive category, laundry detergent, will likely stand out from the others. The current distribution of WTP ratings can be found in figure 3. And the possible removal of outliers is described by figure 4.

Analyses of normality tests still indicated non-normal residuals (Kolmogorov-Smirnov (χ2₌

0.039, p < 0.01), Jarque-Bera test (χ2 _{= 8967.32, p < 0.01) and the Lilliefors test (χ}2 _{= 0.04, p < 0.01))}

for model 4.1. It is observed that the new dataset still contains outliers and therefore decide against removal of outliers. A large majority of the observations that form outliers belong in the detergent category. Not only would the initial removal of these 948 extra outliers distort the findings, but it would not even result in fixing the non-normality of the residuals of the model. Even further removal of outliers may change outcomes more drastically because of distorted results in the detergent category. For this reason, it was decided to continue the analyses without removing any outliers.

Although the data has been found to be not normally distributed, the Gauss-Markov theorem states that an OLS model is still the best linear unbiased estimator as long as the residuals are uncorrelated, are homogeneously distributed and the total of all residuals equals 0. The models already adhere to the RESET-test. To test the latter two issues, multicollinearity and

homoscedasticity will be determined in the following sections.

4.2.3 Multicollinearity

To test the model for multicollinearity, the generalized variance inflation factor (GVIF) scores of the dependent variables are calculated. For single coefficient terms, VIF equals the GVIF score. However,

Figure 4: WTP-rating distribution after further removal of outliers in figure 3

(27)

as the models include several factorial variables, the GVIF can be used to create VIF scores that are comparable across different numbers of parameters (Fox & Weisberg, 2011). Cohen (2003) stated that a VIF score larger than 10 requires transformation of the model due to model distortion. And a VIF score of 4 would require further investigation into the matter. This model distortion is caused by the fact that the variance explained by each explanatory variable can no longer be identified

uniquely. Multicollinearity will therefore cause incorrect model coefficients and significances. The GVIF-scores of the models can be found in table 6. None of the GVIF-scores exceeds the critical score of 4.0. Due to this result, it can safely be assumed that there is no multicollinearity among the explanatory variables. This will allow for accurate interpretation of the regression coefficients and will provide insight on the correctly calculated parameter significance ratings.

4.2.4 Homoscedasticity

Homoscedasticity determines whether the variance of the error-terms are equal across all explanatory variables in the models. Violation of this assumption would distort the findings. Heteroscedasticity is especially prevalent in datasets that use cross-sectional data for estimation (Leeflang et al., 2015). Heteroscedasticity can be tested by using both Bartlett’s (1937) test, Levene’s test or the Goldfeld-Quandt (1965) test. Due to the violation of the normality assumption in 4.1.2, Bartlett’s test is incompatible and cannot be used. Bartlett’s test is less robust and will therefore more easily be influenced by non-normally distributed data. At first, a visual check is done to determine whether Levene’s test or Goldfeld-Quandt test is necessary to determine

heteroscedasticity. The non-standardized residuals and standardized residuals are visualized in a scatterplot on the y-axis versus the fitted/estimated values according to the model on the x-axis. The scatterplots for model 4.1 and 4.2 can be found in appendix 7.5.

Ideally, both plots should visualize a large cloud of residuals with a horizontal fit line in the middle. However, as shown very slightly in appendix 7.5, for the unstandardized residual plot and standardized residual plot, both reference lines seem to be skewed as fitted values increase. This is an indication of heteroscedasticity within the research sample. Whether heteroscedasticity actually exists is tested by using Levene’s test, as this test allows splitting the sample based on another variable. Levene’s test will reassure the doubts that were caused by the scatterplots of the residuals. After calculations, Levene’s test provides inconclusive results after testing for heteroscedasticity for the sample. On the one hand, comparing the data based on educational level (F = 1.02, p = 0.312) or gender (F = 2.06, p = 0.128) does not result in violation of the assumption of homoscedasticity. On

Model 4.1

Model 4.2

Variables GVIF GVIF

Product category 2.01 1.99

Product type 2.03 2.02

Product category * Product type 1.51 1.51

Situation - 1.02

Leading national brand dummy 1.05 -

Non leading national brand dummy

1.05 -

High quality private label dummy 1.05 -

Economy private label dummy 1.05 -

(28)

27 the other hand, splitting the dataset based on age groups (F = 10.63, p < 0.01) and marital status (F = 8.54, p < 0.01) allows the Levene’s test to reject the null hypothesis of homoscedasticity.

Model 4.2 encounters a similar problem. There is a certain increase in residuals as the fitted values increase. This may be an indication of heteroscedasticity. Levene’s test again provides inconclusive results as significant and insignificant results are provided depending on what variable is used to split the sample with. Splitting based on gender (F = 1.02, p = 0.312) and education (F = 2.18, p = 0.113) again indicate no heteroscedasticity across residuals, but splitting the sample based on age groups (F= 10.64, p < 0.01) or marital statuses (F = 8.71, p < 0.01) does the opposite. It is likely that the heteroscedasticity is caused by the older age group. This age group displays a much larger range of residuals than the other age groups do. This deviation from homoscedasticity can be found in figure 5 below. As the oldest age group only consists of 1 participants, it is believed that the observations for this participant will not cause heteroscedasticity on its own due to the large sample size. However, in combination with the 65-80 years old age group, they most likely do. This

collapsing of levels is displayed in appendix 7.6. It clearly visualizes an enlarged distributed of

residuals in comparison to the other age groups. This finding is in line with the slope of the reference line in the scatterplots in appendix 7.5. A similar visual analysis was done for marital status, as displayed by figure 6, but no large differences in residual distribution was observed.

Levene’s test’s result is likely influenced by the fact that splitting the dataset in a higher number of groups to compare variance with will increase the likelihood of finding unequal variances. As both regression weights and the Box-Cox transformation are already used to transform the

Figure 6: Distribution of standardized residuals across age groups

WILLINGNESS TO PAY FOR PRIVATE LABELS AND NATIONAL BRANDS AT HARD DISCOUNT RETAILERS

WILLINGNESS TO PAY FOR PRIVATE LABELS AND NATIONAL

BRANDS AT HARD DISCOUNT RETAILERS

A study on the effects of including premium private labels and national brands in a

hard discount retailer’s assortment

WILLINGNESS TO PAY FOR PRIVATE LABELS AND NATIONAL

BRANDS AT HARD DISCOUNT RETAILERS

A study on the effects of including premium private labels and national brands in a

hard discount retailer’s assortment

Management summary

Preface

Table of contents

1. Introduction

2. Conceptual background

2.1

Price and formats in the retail environment

2.2

Price image in relation to products’ perceived costs

2.3

Different effects of private labels across product categories on

price image formation

2.4

Perceived differences between NB and PL and their marketing

strategies

2.5

Offering national brands at hard-discount retailers

2.6

High quality private label products as a solution?

2.7

Conceptual model

3. Methodology

3.1

Operationalization of constructs

3.2

Data collection

3.3

Online questionnaire

3.4

Participants

3.5

Data cleaning

3.6

Plan of analysis

4. Results

4.1

Model estimation

4.2

Residual analysis

𝑃

=