• No results found

THE ONE WHO KNOWS, PAYS? THE EFFECTS OF PRICE KNOWLEDGE, PRODUCT KNOWLEDGE AND STORE PRICE IMAGE ON THE WILLINGNESS TO PAY AT THE PREFERRED STORE.

N/A
N/A
Protected

Academic year: 2021

Share "THE ONE WHO KNOWS, PAYS? THE EFFECTS OF PRICE KNOWLEDGE, PRODUCT KNOWLEDGE AND STORE PRICE IMAGE ON THE WILLINGNESS TO PAY AT THE PREFERRED STORE."

Copied!
107
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

THE ONE WHO KNOWS, PAYS? THE EFFECTS OF PRICE

KNOWLEDGE, PRODUCT KNOWLEDGE AND STORE PRICE IMAGE

ON THE WILLINGNESS TO PAY AT THE PREFERRED STORE.

A study on the effects of price knowledge, product knowledge and store price image

on the willingness to pay of consumers for supermarket goods at the preferred supermarket in the Netherlands

BY

JESSE LANDSMAN

(2)

The one who knows, pays? The Effects of Price

Knowledge, Product Knowledge and Store Price Image

on the Willingness to Pay at the Preferred Store.

A study on the effects of price knowledge, product knowledge and store price image on the willingness to pay of consumers for supermarket goods at the preferred supermarket in the Netherlands Jesse Landsman Groningen January 2017 Master Thesis MSc Marketing Intelligence

Faculty of Economics and Business - Department Marketing University of Groningen

Supervisor: Dr. J.E.M. Van Nierop

External Supervisor: Dr. Ir. M.J. Gijsenberg

Contact Address: Duindoornstraat 241, Groningen Contact Number: 0640977611

Contact Mail: j.landsman@student.rug.nl

Student Number: S2221519 Date: 16/01/2017

(3)

P

REFACE

Before you lies the thesis “The one who knows, pays?”, which researches the effects of consumer knowledge on the willingness to pay. This thesis has been written to fulfil graduation requirements for the Master Marketing Intelligence at the University of Groningen. I wrote this thesis in the period starting from September 2016 till January 2017. I quite struggled with finding a relevant research question, finding relevant literature and doing the right analyses, but in the end it was a very useful experience.

I am thankful to the people from the Marketing Department of the University of Groningen who were involved in the writing of this thesis. A special thanks goes out to my supervisor Dr. J. E. M. Van Nierop, whose constant helpful guidance and feedback helped me through all the analyses and conclusions. I would also like to thank my family, friends and girlfriend for their moral support.

I hope you enjoy reading this thesis. Jesse Landsman

(4)

M

ANAGEMENT SUMMARY

Price competition is fierce in the sector of supermarkets, calling for constantly optimized pricing schedules, making research on influencing factors of the willingness to pay (WTP) important. Consumer knowledge became a dynamic concept with the rise of the internet, but it is relatively unexplored what impacts these changes in knowledge have on the prices consumers are willing to pay. This thesis focused on the effects of two sides of consumer knowledge on the WTP, which are product knowledge and price knowledge.

Price knowledge is divided into two components, with the differences in the effects of underestimation and overestimation of actual prices being examined. Next to that, the impacts of related concepts product involvement and store price image were analysed. The WTP is measured for the preferred store of consumers in the Netherlands, for the products beer, coffee and laundry detergent. An online questionnaire was deployed, measuring objective and subjective product knowledge, price knowledge by way of price recall, price recognition and deal spotting, price involvement and store price image by way of Likert scales, and a choice-based conjoint analysis was deployed to measure the WTP of the respondents. The survey with a between-subjects design was completed by 283 respondents.

The results of the linear model estimation suggest that no significant direct relationship exists between product knowledge and prices consumers are willing to pay. Product involvement, however, does positively influence the WTP of consumers, but fails to moderate the effect of product knowledge. The differences between underestimation and overestimation of prices are partially significant, with the effects of a reduction in overestimation outweighing the effects of a reduction in underestimation on the prices consumers are willing to pay, which contradicts the expectations of this research. Since the effects of underestimation and overestimation are only partially significant, it cannot be concluded that the accuracy of price knowledge directly influences the WTP of consumers. Additionally, this thesis concludes that a more expensive store price image leads to consumers willing to pay higher prices. The moderating influence of price knowledge on the latter relationship is concluded to be insignificant. A side note has to be made since multicollinearity was found in this research. Therefore, results have to be interpreted with caution.

(5)
(6)

T

ABLE OF CONTENTS

Preface ... 3 Management summary ... 4 1. Introduction ... 8 2. Literature Review ... 12 2.1 Willingness to Pay... 12 2.2 Product Knowledge ... 12 2.3 Product Involvement ... 14 2.4 Price Knowledge ... 15

2.5 Store Price Image ... 19

3. Research Design ... 22

3.1 Product Class selection ... 22

3.1.1 Pre-test ... 22 3.1.2 Pre-test results ... 23 3.1.3 Brand selection ... 24 3.2 Data Collection ... 24 3.3 Measures ... 25 3.3.1 Product Knowledge ... 26 3.3.2 Product Involvement ... 27 3.3.3 Price Knowledge ... 27

3.3.4 Store Price Image... 28

3.3.5 Willingness to Pay ... 29

3.4 Data preparation ... 30

3.4.1 Respondents ... 30

3.4.2 Missing values ... 32

3.4.3 Variable transformations ... 32

3.4.4 Reliability and validity of proposed constructs ... 33

3.5 Data analysis strategy ... 35

4. Results ... 36

4.1 Descriptive statistics ... 36

4.1.1 Product Knowledge ... 36

4.1.2 Product Involvement ... 37

4.1.3 Price Knowledge ... 37

4.1.4 Store Price Image... 39

(7)
(8)

1. I

NTRODUCTION

To reduce the steadily increasing sales figures of discount supermarket concerns Aldi and Lidl, the so-called Big Four of traditionally largest supermarkets in the United Kingdom recently aggressively lowered prices, with that reducing market margins, leading to the closing of least performing stores all over the country (Gale 2016). Price wars in supermarkets are nothing new under the sun, probably any adult in the Netherlands can remember the one Albert Heijn started in October 2003. A sector typified by heavy price competition calls for constantly optimized pricing schedules.

An important concept directly linked to optimizing pricing schedules is the willingness to pay (WTP) of consumers. The assessment of the maximum price a consumer is willing to pay is important input for business decisions regarding product development, competitive strategies and, most importantly, the optimization of pricing schedules (Miller et al. 2014; Wertenbroch and Skiera 2002). Since assessing the WTP of consumers helps optimizing pricing schedules, WTP can be directly linked to the profit equation of organizations (Homburg, Koschate-Fisher and Hoyer 2005). Simply put, a reason for brand managers to assess the WTP of consumers is that if consumers are willing to pay higher prices for their brand, they can ask for these higher prices leading to higher profits (Krishna 1991). Since a relationship between the accurate estimation of the WTP and profitability seems to be present, research is required on influencing factors of the WTP.

(9)

for consumers with lower levels of knowledge. However, Rao and Sieben (1992) questioned the generalizability of their results and stated that it is feasible that found effects could vary across product categories.

In the same article, Rao and Sieben (1992) suggested a possible moderating effect of involvement on the relationship between product knowledge and price acceptability. Involvement has been an important concept in the last few decades in the marketing industry, with product involvement being of great value to consumer theories and the development of these theories (Goldsmith and Emmert 1991). Product involvement has been proven to be of influence on consumer behaviour, such as consumers’ information search and processing, with highly involved individuals being more driven to search and process product information (Warrington and Shim 2000), relating this concept closely to product knowledge. Considering the importance of product knowledge and product involvement in consumer behaviour research and to end a part of the doubts Rao and Sieben stated in 1992, this research examined whether a similar relationship exists between product knowledge and the WTP for fast-moving consumer goods, namely supermarket goods, and whether a moderating effect exists of product involvement on aforementioned relationship. Since Campbell, DiPietro and Remar (2014) found a positive significant effect of product involvement on the WTP, it is interesting to test whether this conclusion holds for non-food products and, for that reason, the direct relationship between product involvement and the WTP is examined as well.

(10)

A concept closely related to price knowledge is store price image. Price images of stores have become a focus point in the retail sector over the last few decades, with the importance of this concept being confirmed by the results of a Nielsen study that came out in 2008, with most people stating that good value for money determines their store choice, with 70% of those consumers expressing that it does not matter whether a store is actually cheaper in comparison with competitors, or only characterized as relatively cheaper (The Nielsen Company 2008). Hamilton and Chernev (2013) stated that consumers, with regards to purchase decisions, are not only influenced by actual in-store prices, but also by their perception of the store’s prices. The importance of store price images became clear in the Netherlands when in October 2003 supermarket concern Albert Heijn started the largest price war to date in the Netherlands, attempting to improve their price image (Van Heerde, Gijsbrechts and Pauwels 2008). The research of van Heerde, Gijsbrechts and Pauwels (2008) indicated the importance of a favourable store price image in a price war, concluding that having an unfavourable price image in a price war could very well lead to losing customers in the long run. With ongoing heavy price competition between supermarket concerns in the Netherlands, store price images are ever so relevant to the retail sector. Hence, this research aimed to examine whether the price image of a store actually also influences the prices consumers are willing to pay for products at that same store. The market of grocery retailers is typified by consumer behaviour consisting of a loyal visit once a week to the preferred supermarket, in which more time is spent than any other visits (Abubakar, Mavondo and Clulow (2001). Creusen et al. (2008) moreover stated that so-called “one-stop” shopping is popular amongst Dutch grocery shoppers, indicating the vast majority of groceries being bought in a single visit to the preferred supermarket. Since supermarket preference seems to correlate with supermarket buying behaviour, it is interesting to research aforementioned effects for the preferred supermarket of the consumer. Therefore, this research focused on influencing factors on the willingness to pay at the preferred supermarket of the consumer.

The aforementioned hypothesized relationships led to the following main research question:

How does consumer knowledge influence the willingness to pay of consumers for supermarket goods at the preferred supermarket?

This main question was answered through the following sub-questions:

1. How does the accuracy of product knowledge influence the willingness to pay, and what are the effects of product involvement in this relationship?

(11)
(12)

2. L

ITERATURE

R

EVIEW

Relevant articles are reviewed in this section of the thesis. The willingness to pay is defined, which is the outcome variable of this thesis. The WTP is followed by the conceptualizing and the expected effect of product knowledge. Product involvement is closely related to product knowledge, thus next described is the definition and expected effects of product involvement. Price knowledge is the other side of consumer knowledge relevant to this thesis, which conceptualization and hypothesized effects are next depicted. Following that, store price image is described and hypothesized, a concept closely related to price knowledge. All aforementioned concepts and hypothesized effects are represented in the conceptual model as can be found concluding this chapter.

2.1

W

ILLINGNESS TO

P

AY

The willingness to pay has been defined as the maximum price a person is willing to pay for a certain product or service (Krishna 1991), also known as the reservation price. Homburg, Koschate-Fisher and Hoyer (2005) defined the WTP as the assessment of a consumer’s value assigned to a consumption or usage experience expressed in monetary units. The WTP could also be conceptualized as the upper bound of the price acceptability range (Le Gall-Ely 2009). The WTP allows an individual to reveal the financial value of their judgment of a product’s value (Le Gall-Ely 2009).

2.2

P

RODUCT

K

NOWLEDGE

(13)

actual knowledge stored in memory. Moreover, subjective and objective knowledge influence consumers’ information processing in different ways. Cowley and Mitchell (2003) agreed with the latter approach and adopted both subjective and objective knowledge measures. Carlson et al. (2009) concluded that using measures of subjective knowledge as a stand-in for revealing objective knowledge is more applicable for durable, hedonic, public, luxury and search goods. In line with Rao and Monroe (1988) and Rao and Sieben (1992), this research defined product knowledge as the amount of factual information stored in a consumer’s memory combined with the consumer’s perception of this product knowledge.

(14)

a higher WTP for the product, compared to individuals with lower knowledge levels. If a consumer were to increase his or her product knowledge, the uncertainty with regards to the product would reduce, leading to a decreased notion of missing information, which improves the value perceptions of a product. Higher product evaluations cause a higher WTP, which led to the following hypothesis:

H1: The price that consumers are willing to pay increases with their accuracy of product knowledge.

2.3

P

RODUCT

I

NVOLVEMENT

Goldsmith and Emmert (1991) defined product involvement as the attitudinal affection and excitement consumers hold towards product classes. Zaichkowsky (1985) viewed product involvement as the personal relevance of a product to the consumer, based on their personal demands, principles and interests. Warrington and Shim (2000) stated that product involvement is comparable to ego involvement, which means that involvement is high when an object or product is related to the character and principles of the consumer’s self-concept. A distinction can be made between situational and enduring involvement, where situational involvement reflects a more short-term interest and enduring involvement a continuing interest in the product category (Warrington and Shim 2000). While all definitions are comparable in ways of conceptualizing product involvement as a personal connection to the product, the definition by Zaichkowsky (1985) is often applied by other researchers (e.g. Lin and Chen 2006, Warrington and Shim 2000), which led to the adoption of their definition for this report.

Consumers that display higher product involvement reveal more active information search behaviour and are more inclined to participate in a higher extent of decision-making processes with regards to the product category (Warrington and Shim 2000). Zaichkowsky (1985) stated similar relationships, after reviewing multiple articles she concluded that consumers displaying higher product involvement reveal more active information search behaviour with regards to brands and more product characteristics comparing. Lin and Chen (2006) found a positive relationship between product involvement and information search intention. Chaudhuri (2000) confirmed that a relationship between product involvement and information search exists.

(15)

obtained this with different reasons. When the product knowledge is obtained through higher involvement, and thus because of personal demands, principles and interests, their product knowledge might influence their WTP even more positively. Moreover, Lin and Chen (2006) confirmed the existence of the moderating effects of product involvement, concluding in their article that product involvement positively moderates the relationship between product knowledge and purchase intention. It was expected that product involvement positively influences the effect of product knowledge on the WTP, therefore it was hypothesized:

H2: Product involvement strengthens the relationship between product knowledge and the willingness to pay.

A higher product involvement leads furthermore to consumers being able to differentiate brands better, which leads to these consumers stating more brand preferences (Zaichkowsky 1985). Lin and Chen (2006) stated that a significant positive relationship between product involvement and purchase intention is present. Liang (2012) discovered that product involvement stimulates impulse buying behaviour, which could be associated with a reduced focus on prices. Consumers displaying higher product involvement therefore seem to be having a more explicit preference for certain brands, are more affected by the product itself and therefore are more willing to buy these products, which both point towards a higher value for the products for which product involvement is higher. A higher value can be reflected in a higher WTP for a product. Campbell, DiPietro and Remar (2014) proved that a positive significant relationship between product involvement and the WTP is existent for local foods. Lichtenstein, Bloch and Black (1988) added that consumers that display higher product involvement can be less involved with prices, and therefore reveal a higher WTP. Additionally, since product involvement is proven to lead to higher product knowledge, which is expected to positively influence the WTP of consumers, it is anticipated that a similar positive relationship between product involvement and the WTP exists. Therefore, it is hypothesized:

H3: The price that consumers are willing to pay increases with their product involvement.

2.4 P

RICE

K

NOWLEDGE

(16)

Dickson 1991; Vanhuele and Drèze 2002). Research to different drivers of price knowledge furthermore gained popularity, with varying results of the influence of decision characteristics (Dickson and Sawyer 1990), satisfaction (Homburg, Koschate-Fisher and Wiegner 2012) and consumer and product characteristics (Estelami 1998; Krishna, Currim and Schoemaker 1991; Vanhuele and Drèze 2002).

Studies like Dickson and Sawyer’s research (1990) (e.g. Le Boutillier, Le Boutillier and Neslin 1994) defined price knowledge as the ability to recall prices after revealing purchase intention by way of item choice. More recent studies critiqued this way of conceptualizing price knowledge, pointing out that price knowledge is more than the remembering of price information (Monroe and Lee 1999) and the mere focus on short-term memory of Dickson and Sawyer’s definition (Vanhuele and Drèze 2002). Monroe and Lee (1999) indicated that not only explicit memory, or the conscious remembering of an event, but also implicit memory, which is the information stored in the mind of a consumer without the individual being able to consciously recall this information, influence product and price evaluations as well as purchase decisions. This translates to price knowledge consisting of two components: explicit price knowledge, which is the past prices or price information that are consciously remembered and implicit knowledge, which is the unconscious knowing of price information (Homburg, Koschate-Fisher and Wiegner 2012). Measures consisting of only price recall or price estimation fall short since consumers might have price information stored in memory which they simply cannot access consciously (Monroe and Lee 1999).

(17)

processing are unlikely to lead to price awareness, but are likely to update price knowledge (Monroe and Lee 1999). The study of Dickson and Sawyer (1990) focused according to Vanhuele and Drèze (2002) mostly on short-term memory, while the price knowledge of consumers naturally is stored in the long-term memory. By way of asking customers to recall prices moments after item choice, the outcome is more influenced by the attention paid to prices during shopping instead of the actual price knowledge brought to the store. Price knowledge is therefore defined as the combination of conscious and unconscious prices and price information stored in the memory of consumers, in line with Homburg, Koschate-Fisher and Wiegner (2012).

Another important issue regarding price knowledge and reference prices, is the issue of overestimation and underestimation. Overestimation, in other words having a reference price higher than actual prices, therefore the actual price being lower than expected, leads to consumers having a more positive perception with regards to a certain price and thus having a higher WTP. In the same sense, underestimation of the same actual price leads to a lower WTP. A higher accuracy of price knowledge leads to more accurate reference prices, thus reducing overestimation and underestimation. The difference in the effects of a decrease in overestimation and a decrease in underestimation can be predicted by the theory of loss aversion. This theory is applicable to overestimation and underestimation according to Kalyanaram and Winer (1995). The theory of loss aversion depicts that the impact of a value loss, compared to a certain reference point, is greater than the impact of a value gain, compared to this same reference point (Thaler 1985). In other words: “losses loom larger than corresponding gains” (Tversky and Kahneman 1991, p. 1039). This results for the value function in a concave effect for gains and a convex effect for losses (Thaler 1985). A visual representation of this effect was developed by Tversky and Kahneman (1991), as can be seen in figure 1 on the next page.

Figure 1: visual representation of loss aversion on the value function (Tversky and Kahneman 1991)

(18)

and Winer 1995). After reviewing relevant literature, Kalyanaram and Winer (1995) came to the conclusion that evidence could be found for the empirical generalization of consumers being more sensitive to underestimation compared to overestimation. If a person that underestimated the actual price were to increase his or her price knowledge, in other words, develop a more accurate reference price, decreasing underestimation, the increase in this consumer’s WTP is estimated to be larger than the decrease in WTP of a consumer gaining price knowledge who overestimated actual prices. Expected relationship is visually represented in figure 2, and led to hypothesis 3:

Figure 2: visual representation of loss aversion with regards to reference prices and the WTP.

H4: The effect of an increase of price knowledge on the willingness to pay is stronger for underestimation compared to overestimation.

(19)

H5: The price that consumers are willing to pay increases with their accuracy of price knowledge.

2.5

S

TORE

P

RICE

I

MAGE

Store price image (SPI) has been defined as the perception of consumers with regards to how cheap or expensive a certain retailer is (Lourenço, Gijsbrechts and Paap 2015). A lower SPI is a cheaper, a more favourable price image. A higher SPI is a more expensive, less favourable price image. Hamilton and Chernev (2013) defined this concept as the consumers’ judgment of the overall price level of a store. The initial, first store price image is not formed by actual prices, but developed through non-price cues that consumers expect to correspond with non-price levels of a store, such as the location and the interior of the store, as well as the promotional activities and the selection of products the store offers (Hamilton and Chernev 2013; Lourenço, Gijsbrechts and Paap 2015). This initial price image of a store is updated during store visits and actual encounters with store prices, which happens consciously or accidental (Lourenço, Gijsbrechts and Paap 2015), similar to the updating of price knowledge. Store price image is in this way connected to price knowledge, since the procedure of processing price information and acquiring price knowledge could also influence the price image of the store (Dickson and Sawyer 1990). Moreover, the price image of a store is similar to a reference price in that both have an impact on price perception (Hamilton and Chernev 2013). However, SPI differs from price knowledge in that consumers determine a store’s price image mostly through a mixture of price-relevant and non-price cues, instead of basing it only on actual store prices (Hamilton and Chernev 2013).

(20)

individuals, it is expected that this will have a negative effect on the WTP of consumers visiting these stores. On the other hand, individuals are less likely to postpone purchase to compare prices between stores when shopping in a store with lower SPI, since the perceived possibility of finding a lower price is small (Hamilton and Chernev 2013). This suggests consumers being more likely to accept prices charged by stores with a lower SPI, hinting at a possible higher WTP. Store price image furthermore influences the price judgment of consumers. According to Hamilton and Chernev (2013), a group of references stated the relationship between SPI and price evaluations as consumers judging prices in line with the SPI, while other articles added that consumers modify their reference prices considering the store’s price image. Consumers lower their reference price for stores with a lower SPI, and modify reference prices for stores with a higher SPI upwards. This could lead to a more positive perceived price of the same product in a store with a higher SPI compared to a store with a lower SPI, suggesting a higher WTP for stores with a higher SPI. Aforementioned references tend to suggest a positive influence of a more expensive price image on prices consumers are willing to pay, which led to the following hypothesis:

H6: The price that consumers are willing to pay increases with the store price image.

Hamilton and Chernev (2013) furthermore expressed that recent articles stated a difference in how SPI influences the price judgment of consumers with precise reference prices compared to consumers without a clear-cut reference price. While a consumer with well-defined reference prices is able to adjust their reference prices in line with the stores SPI, leading to less favourable price judgments for stores with a lower SPI and more favourable price judgments for stores with a higher SPI, consumers not possessing these clear-cut reference prices in mind tend to evaluate prices in line with the SPI, leading to more positive price perceptions for stores with lower SPIs (Hamilton and Chernev 2013). More precise reference prices are caused by more accurate price knowledge, which led to the expectation of a positive moderating effect of price knowledge. Therefore, it was hypothesized:

H7: The accuracy of price knowledge strengthens the relationship between store price image and the willingness to pay.

(21)
(22)

3. R

ESEARCH

D

ESIGN

In this section, the design of the research is described. The process of selecting appropriate product classes used for testing is defined, followed by a description of the data collection. Adopted measures are illustrated next, followed by an explanation of data preparation and the data analysis strategy.

3.1

P

RODUCT

C

LASS SELECTION

In line with prior research, two criteria were used to select the most appropriate product categories for this study. The first criterion was that prices within the product category needed to differ (Lichtenstein, Bloch, Black 1988) and second, levels of product involvement had to differ between consumers (Lichtenstein, Bloch, Black 1988; Zaichkowsky 1985). Differences in price can easily be observed, variation in product involvement is less obviously detected. Therefore, a pre-test was deployed to select the optimal product categories for this research.

3.1.1 Pre-test

(23)

confirm the expected higher variability in product involvement for the products beer, laundry detergent and beer in comparison with the products yoghurt, mineral water and toothpaste.

Zaichkowsky (1985) developed a semantic differential scale called the Personal Involvement Inventory. The original scale consisted of 20 word groups that had to be rated using a 7-point scale. This 20 item scale is often attempted to be reduced, for example by Zaichkowsky herself (1994). Reduced forms of the 20-item scale are still one of the most commonly used (e.g. Lin and Chen 2006; Prendergast, Tsang and Chan 2010) scales to measure product involvement. Mittal (1995) reduced Zaichkowsky’s (1985) initial scale to five items, which displayed satisfactory validity and simplicity to use (Mittal 1995). Mittal’s (1995) reduced form of Zaichkowsky’s (1985) scale has been used recently as well (e.g. Atakan, Bagozzi and Yoon 2014; Li and Lo 2015), which led to the adoption of this method. Word groups used for this pre-test were important/unimportant, of no concern to me/of concern to me, means a lot to me/means nothing to me, matters to me/does not matter to me, significant/insignificant.

The study sample of this pre-test consisted of 25 consumers who were contacted via Facebook. Of the total of 25 participants that started the questionnaire, 20 respondents (80%) successfully completed it. Since the study was concerned with consumer behaviour in Dutch supermarkets, all participants had the Dutch nationality and the questionnaire was consequently translated to Dutch. A within-subjects design was adopted, all participants scored all six products on the five-item scale. Back-translation was used to examine the accuracy of the translation and to expose potential misunderstandings (Brislin 1970), which is a commonly adopted technique in survey research (Douglas and Craig 2007). The back translator, with an academic level of English language, translated the Dutch version of the questionnaire and after no major disagreements and some discussion the proposed translation was accepted. The second goal of this pre-test was to confirm the validity and reliability of the 5 item scale by performing a factor and reliability analysis. The questionnaire is illustrated in appendix 1.

3.1.2 Pre-test results

(24)

3.1.3 Brand selection

To enhance the possibility that participants regularly purchase or consider purchasing the selected products and to enhance the simulation of real shopping decisions, well-known brands were selected (Urbany and Dickson 1991), in combination with popular packing sizes.

For the product beer, regular sized beer crates, holding 24 bottles containing 0.3 litre of regular, pilsner beer were selected, because pilsner is the most popular store-bought type of beer in the Netherlands (De Jongh, Van Teeffelen and De Kruijk 2016). Since supermarket sales per brand are not publicly available, brands were selected on the basis of revenue figures in the hospitality sector in the Netherlands. Heineken, Grolsch and Hertog Jan ranked first, second and fourth with regards to hospitality revenue, which consequently were selected. Jupiler, accountable for the third largest revenue in the hospitality sector, was not suited for this survey since their package size is divergent. The product coffee was specified as ground coffee, which is according to the research of coffee producer DE one of the most sold types of coffee, with 38% of coffee sales in the Netherlands being ground coffee (DE 2013). According to the same source, Douwe Egberts’s quick filter ground coffee is the single most consumed coffee in the Netherlands. The most common package size of quick filter ground coffee in supermarkets is 500 grams, of which brands Douwe Egberts, Kanis & Gunnink and Van Nelle were selected since their common producer DE displays by far the largest revenue in the hospitality sector in the Netherlands (DatLinq 2014).

The most popular form of laundry detergent in the Netherlands is liquid detergent, of which the concentrated version was one of the bestselling variations in the Netherlands in 2009 (Distrifood 2009). Distrifood mentions brands Robijn, Ariel and Omo as successful in this subtype of laundry detergent, which was the justification for selecting those brands. No data was available whether liquid detergent for coloured, white or black clothes is more popular. For this research, the choice was made to use the coloured version of liquid detergent. All liquid detergent package sizes differed slightly, therefore respondents were asked to assume a common package size that is usable for 20 washloads.

3.2

D

ATA

C

OLLECTION

(25)

product characteristics online. To avoid this, the subjects were asked to fill in honest answers, only concentrate on the questionnaire when answering questions, and were assured that data would be anonymously stored, meaning results could not be tracked back to a certain person. Respondents were assured that a higher or lower product knowledge or price knowledge had absolutely no influence on their odds to win. Additionally, the time used was tracked, leading to the possibility to filter out answers that displayed outlier values in answering time. The full questionnaire consisted of 7 introduction questions, followed by 27 questions for each of the three products, concluded with 3 final questions. To reduce possible fatigue and loss of motivation and concentration amongst respondents, a between-subjects design was adopted for this survey, with each consumer filling in the questionnaire for 2 of the 3 products, reducing the questionnaire length from 91 to 64 questions. Since the Dutch consumer market is researched in this thesis, only Dutch consumers were targeted and the questionnaire was consequently translated to Dutch. The questionnaire was distributed and collected by the researcher.

3.3

M

EASURES

(26)

Figure 4: structure of the questionnaire, every respondent goes through two of the three possible pathways.

3.3.1 Product Knowledge

Two measures were applied to measure product knowledge at the product class level, starting with one open question and three multiple choice questions aimed at revealing objective product knowledge, followed by one subjective knowledge question. Carlson et al. (2009) expressed that measuring objective knowledge with closed-ended questions could facilitate guessing of correct answers, which might lead to a considerable bias. Prior research, however, (e.g. Cowley and Mitchell 2003; Lee and Lee 2009; Rao and Sieben 1992) frequently used closed-ended questions to measure objective knowledge and time constraints stopped this research from evaluating hundreds of open answers. To reduce the possible bias of using only closed-ended question, the open-ended question as used by Cowley and Mitchell (2003) was added to the objective product knowledge measure, asking the respondent to recall brands of a certain product. Unfortunately, to the researcher’s best knowledge, there are no objective product knowledge scales developed for supermarket goods. Therefore, a new scale was developed, based on Rao and Sieben’s (1992) product knowledge scale. The scale developed by Rao and Sieben (1992) consisted of 17 items, which were each rated on importance by experts. Consequently, three closed questions were developed for this research, based on Rao and Sieben’s most weighted items, which can be found in appendix 4. One question about the ingredients of the product, one question about the durability of the product and one question about the consumption of the product were added to the objective product knowledge scale, as illustrated in appendix 5.

(27)

questions, aimed at the self-perception of knowledge, familiarity and usage levels. Biswas, Biswas and Das (2006) adopted four questions, aiming to reveal self-reported knowledge, expertise, familiarity and experience levels. To keep the questionnaire compact, subjective product knowledge was measured by asking respondents to rate themselves on a 7-point scale reaching from “not very knowledgeable” to “very knowledgeable”, as used by Cowley and Mitchell (2003). The reason for adopting this single question is that this question captures, in essence, the self-reported, subjective product knowledge of a consumer.

3.3.2 Product Involvement

Besides the scale of Mittal (1995) still being used frequently by researchers, the reliability analysis of the pre-test displayed positive results, leading to the adoption of this scale of product involvement for the main survey. The word groups used for this scale can be found in the pre-test section.

3.3.3 Price Knowledge

Since price knowledge is more than just the recalling of an exact price after item choice, a more comprehensive scale was needed to measure this type of knowledge. Vanhuele and Drèze (2002) acknowledged the differences between explicit and implicit knowledge and developed a methodology that accounts for both these different forms of price knowledge, as well as taking into account both short-term and long-term memory. In line with Vanhuele and Drèze (2002), three different constructs measuring three forms of price knowledge were used for this research: price recall, which is the knowing of the exact price of a product, price recognition, which is the ability to identify the price they are used to see and deal spotting, which is the ability to assess whether a presented price is within the normal price range. This combination of three different questions assures a more accurate and complete picture of the individual’s actual price knowledge (Vanhuele and Drèze 2002) and is adopted by recent research as well (e.g. Jensen and Grunert 2014).

(28)

For all three measures, price knowledge was consequently operationalized as the price estimation error, in line with Eisenhauer and Principe (2009). The price estimation error is calculated as the proportional difference between the estimated price and the actual price. A higher value for any of the three measures of price knowledge consequently represents a lower price knowledge and vice versa. To measure price recall, pictures of a product were followed by the question “What do you think is the price of this brand of [product] at [preferred supermarket] today?”. Respondents were asked to give an as accurate as possible price estimation. For the second measure of price knowledge, price recognition, pictures of the products were presented, this time accompanied with a selection of 6 different prices, of which the respondent was asked to indicate the normal price at their selected favourite supermarket. Alongside the actual price, five alternative prices that differed between 5 and 20 percent were displayed, with lower priced products having a relatively wider range of prices than higher price products. The prices of alternatives all simulated credible prices, in other words, prices that are actually used in supermarkets for the type of product (e.g. €10.49 instead of €10.41), reducing the possibility of correct guessing (Jensen and Grunert 2014). To avoid selection bias, the location of the actual price was equally distributed across questions (Jensen and Grunert 2014).

The third and final measure, deal spotting, involved an image of a certain product followed by a sequence of hypothetical prices, one at a time, 2 different prices per brand. Consumers were asked to indicate whether the displayed price was higher or lower than the normal price of the product. The two prices were randomly selected out of 4 options, varying -5%, -10%, +5% and +10% of the actual price of the product, as a combination of the variation used by Jensen and Grunert (2014) and Vanhuele and Drèze (2002). The range of prices used for deal spotting was smaller than price recognition, to avoid correct guessing on the basis of choice options of the price recognition measure. The prices were rounded, that is €10.50 instead of €10.49, to avoid repetition of prices. The order of the questions for price recognition and deal spotting were randomized. Furthermore, the order of prices was randomized across questions, avoiding respondents to predict the direction of next prices (Vanhuele and Drèze 2002). Actual in-store prices are illustrated in appendix 6, which were regularly checked during questionnaire distribution to detect price changes. Actual prices remained constant during the distribution of the questionnaire.

3.3.4 Store Price Image

(29)

being the store, the concrete attribute being the price image. After considering these references, the question as used by Lourenço, Gijsbrechts and Paap (2015) was adopted to measure store price image. Additionally, it was made clear that prices in this question included discounts and promotions.

3.3.5 Willingness to Pay

Popular ways to estimate the WTP of consumers are, among others, the choice-based conjoint analysis or the direct questioning method. Miller et al. (2014) stated that ordinary choice-based conjoint analyses and the direct questioning method could suffer from hypothetical bias. An option to prevent this hypothetical bias is adopting an incentive aligned conjoint analysis, but unfortunately, it was not within the reach of this research to adopt this method. However, ordinary choice-based conjoint analyses are still able to represent true demand sufficiently, and lead to appropriate pricing decisions (Miller et al. 2014), leading to this method still being useful for this research. Therefore, an ordinary choice-based conjoint was adopted in this research to measure the WTP of consumers.

The ordinary choice-based conjoint consisted of three attributes with each three levels. The choice of three levels per attribute was made because it is frequently used in other research, and by keeping the number of levels per attribute constant, the number-of-levels effect was avoided (Eggers et al. 2015). Attributes brand, supermarket and price were selected for the conjoint analysis. A no-choice option was added to these three attributes, which contributed to the realism of the choice situation (Eggers et al. 2015).

(30)

price of €5.49. A no-choice option was included for all choice sets to allow for WTP measurement. All attribute levels are represented in table 1.

A fractional factorial design was adopted to limit the number of questions, avoiding fatigue amongst respondents. To assess the number of choice sets per individual, it had to be taken into account that a higher number of choice sets leads to higher reliability, but at the same time induces fatigue amongst respondents, which could lead to errors (Eggers et al. 2015). Eggers et al. (2015) stated that a design with six to ten choice sets could already facilitate a healthy estimation of consumer preferences. Furthermore, after reviewing several articles applying CBC analyses, Eggers et al. (2015) came to the conclusion that most articles made use of a number of choice sets between 5 and 9. For this research, 9 choice sets per product per person were used. A translated version of the used questionnaire can be found in appendix 9. A balanced and orthogonal fractional design was developed with the use of Preference Lab software.

Table 1: attribute levels of the choice-based conjoint analysis.

3.4 D

ATA PREPARATION 3.4.1 Respondents

Of the 481 consumers that started the survey, 283 completed all questions (58.8%). The initial gender distribution of the sample, as represented in table 2, revealed an underrepresentation of females compared to the actual shopper population in the Netherlands in 2014 (EFMI Business School / CBL 2014). Additionally, halfway through the distribution of the survey, the observation was made that the relative share of respondents above 54 years old was slacking. Quota sampling was used to improve this percentage, which led to a conclusive share of 20.5% of respondents aged 55 or above. As seen in table 2, consumers aged 55 or above were still underrepresented compared to the actual population. Abovementioned under- and overrepresentation led to the adjustment in weights of the cases, which led to a representative sample with regards to age and gender, as represented in table 2.

Brand Supermarket Price

Detergent Coffee Beer For all products equal Detergent Coffee Beer

Level 1 Ariel Douwe

Egberts

Grolsch Albert Heijn €5.59 €4.49 €13.49

Level 2 Omo Kanis & Gunnink

Heineken Jumbo €6.19 €4.99 €14.99

Level 3 Robijn Van Nelle Hertog Jan

(31)

Table 2: representativeness of age and gender of the sample.

After the distribution of age and gender was made in line with the actual population, the distribution of gross income, graduation degree and household composition was examined. The income distribution, as represented in table 3, is roughly in line with the actual income average, being €23,200 in 2014 (Centraal Bureau Voor De Statistiek 2015). Graduation degrees of the sample were quite in line with the population of the Netherlands (Centraal Bureau Voor De Statistiek 2016b), with an underrepresentation of primary school degrees, and an overrepresentation of university (Dutch: HBO and WO) graduates. Household composition was moderately well represented in the sample, single person households being underrepresented and households consisting of 3 persons or more being overrepresented (EFMI Business School / CBL 2014). Gross income, graduation degrees and household composition are of less interest to this research, therefore no further weighting of cases is applied.

Table 3: distribution of gross income, graduation degree and household composition of the respondents

Representativeness

Gender Initial After weighting Actual

Male 42% 29% 29% Female 58% 71% 71% Age 18-34 45% 24% 21% 35-54 35% 41% 41% 55+ 20% 35% 35%

Gross Income Sample Actual

(32)

The combination of the products beer and laundry detergent was displayed most often in this research (39.0%), followed by the combination of beer and coffee (30.7%), and the combination of coffee and laundry detergent (30.4%). Chi-square and one-way ANOVA tests were performed to test for the homogeneity of the samples for the three different products, which is crucial to the validity of a between-subjects design (Jensen and Grunert 2014). The results, which can be found in appendix 7, indicate no significant differences in the samples for product combinations beer and laundry detergent, beer and coffee and laundry detergent and coffee, which suggests that a sufficient comparability of the samples can be concluded (Jensen and Grunert 2014).

3.4.2 Missing values

From the 283 completed surveys 10 responses were deleted, since these cases displayed outlier values for time used to complete the survey. All brands that were not qualified as an actual brand of the product, for example private labels or types of the product, were deleted from the product knowledge open question items. Unfortunately, due to a bug in the Qualtrics software, the price recall question displayed missing values for more than two-thirds of the sample. Due to this major decrease in sample size, it was decided not to incorporate the price recall question in the data analysis. Price knowledge of consumers consequently consisted of two components in the analysis, namely price recognition and deal spotting.

3.4.3 Variable transformations

The open question measuring objective product knowledge on the basis of the number of brands that consumers could recall was transformed into a score between 1 and 10, with one point for each recalled brand. The values for the closed questions measuring product knowledge were each transformed into values ranging from 0 to 5, with a minimum score for no correct answers and a maximum score for all correct answers selected, to assure a scoring in line with the open objective product knowledge question. For the same reason, the range of values for the subjective product knowledge question was adjusted to a minimum score of 0 and a maximum score of 10.

(33)

compared on predictive quality in chapter 4, which led to the adoption of the preferred option in the linear model.

3.4.4 Reliability and validity of proposed constructs

To assess the validity and reliability of the constructs proposed in subsection 3.3 of this thesis, a factor analysis was executed, by way of a principal component analysis, which is the most relevant method for data reduction (Hair et al. 2009). It has to be mentioned that the price knowledge measures will not be combined in a single construct, but rather will be estimated separately, which is in line with Jensen and Grunert (2014) and Vanhuele and Drèze (2002). The Varimax orthogonal rotation method was deployed in the factor analysis, which is the most common used method for rotation and the favoured approach when data reduction by way of a smaller number of variables is intended (Hair et al. 2009).

The results of the factor analysis, as illustrated in table 4, indicate a sufficient score for the KMO test of sampling adequacy and Bartlett’s test of sphericity for all three products, which means that it is allowed to perform a factor analysis on the data. After deleting the closed questions aimed at revealing product knowledge by way of consumption and durability knowledge of a product, the principal component communality scores for all items exceeded the recommended threshold of .50 (Hair et al. 2009). Total variance explained by the two proposed components exceeded the threshold of 60% (Hair et al. 2009). However, the rotated component matrix displayed significant cross loading (>.50, Hair et al. 2009) for two of the three products for product knowledge items.

Table 4: results of the factor analysis for product knowledge and product involvement, with rotated factor cross-loadings on the other construct in parentheses.

Coffee Laundry Detergent Beer

KMO, Bartlett .84 .000 .81 .000 .89 .000

Variance explained by 2 components

75.7% 64.8% 80.2%

Product Knowledge Communalities Rotated factor loadings

Communalities Rotated factor loadings

(34)

While other criteria are met, cross-loadings indicate a possible threat to the reliability of the proposed components. Therefore, a reliability analysis was performed, which results are illustrated in table 5. For the adapted constructs of product knowledge and product involvement, the Cronbach’s alpha coefficient was obtained. The proposed product involvement construct was approved, consistent with the pre-test results. Furthermore, these analyses indicated a possible increase in Cronbach’s alpha if the closed question item measuring objective product knowledge was deleted from the product knowledge construct. For this newly proposed construct, consisting of 2 items, the Spearman-Brown split-half coefficient was obtained, indicating an improvement for two out of three products. However, for both coffee and laundry detergent, the reliability did not meet the threshold of acceptability of .70 as advised by Hair et al. (2009).

Table 5: Reliability analysis output for proposed components, with the number of items in parentheses.

Since the factor analysis and reliability analysis both cannot guarantee a sufficient validity and reliability of the proposed product knowledge construct, it was decided to not combine product knowledge measures, but to keep the measures separately. Product knowledge was consequently measured by the open question aimed at objective product knowledge and the self-reported, subjective knowledge measure, since both items display sufficient communalities and are used in articles on product knowledge.

Another conclusion that is obtained from the factor analysis results is the possible danger of multicollinearity between the two product knowledge items and the product involvement construct, indicated by cross-loadings. Especially the subjective product knowledge item was suggested to be at risk of displaying high correlations with other variables, since both this item and the product involvement construct are self-perceived variables, which moreover are related, indicated by the hypothesized interaction effect. To assess whether a significant multicollinearity existed between these variables, a variance inflated factor test was executed during model estimation.

Coffee Laundry Detergent Beer

(35)

3.5

D

ATA ANALYSIS STRATEGY

(36)

4. R

ESULTS

To get a better grasp on the collected data, a few descriptive statistics are presented in the first section of this chapter. The first section is followed up by the model estimation, consisting of the calculating of the WTP values, a test whether pooling is allowed and the estimation of the final model itself. This chapter is concluded by the acceptance and decline of hypotheses.

4.1

D

ESCRIPTIVE STATISTICS 4.1.1 Product Knowledge

Product knowledge is measured by two items representing two elements of product knowledge, an objective and a subjective component. When looking at the average scores for these product knowledge items, as illustrated in figure 5, it can be concluded that these average scores vary quite heavily across product categories. For the product coffee, the self-perceived product knowledge (measured by the subjective product knowledge question) seems to stroke quite well with the actual or objective product knowledge. On the other hand, the actual product knowledge for the product beer seems to outperform the self-perceived product knowledge. The difference in knowledge for the product beer might be explained by respondents underestimating their product knowledge with regards to beer, or the single item measuring objective product knowledge not representing the total product knowledge sufficiently. Additionally, it can be concluded that people estimate their product knowledge to be quite low, since the average subjective knowledge score ranges from 3.29 to 3.58 on a scale from zero to ten. Product knowledge seems to be the highest for the product beer, since both the objective and the subjective knowledge averages are the relative highest amongst the products.

Figure 5: Average objective and subjective product knowledge accuracy per product per brand, on a scale from 0 to 10. 3,65 4,37 6,67 3,52 3,29 3,58 -1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00

Coffee Laundry Detergent Beer

Average score of product knowledge per

product

(37)

4.1.2 Product Involvement

The average involvement levels of respondents confirm what one could expect, since the sample consists of 71% females and 29% males. The average involvement for the product laundry detergent, traditionally considered to be a more feminine product, is 4.90 on a 7-point scale, while the product beer, which could be considered a more masculine product, displays an average involvement of 2.75 on a 7-point scale. Indeed, the conclusion of the one-way ANOVA test is that males and females significantly (p<.01) differ with regards to involvement for the products beer and laundry detergent. The average involvement on the 7-point scale of the product coffee is 4.38. Standard deviation is the highest for the product beer (2.00), followed by coffee (1.95) and laundry detergent (1.43).

4.1.3 Price Knowledge

As mentioned before, the price recall question was removed from the price knowledge construct because of a large amount of missing values. However, the values that were recorded do hold information. These values are represented in table 6, and will briefly be described. Average recalled prices seem to differ just slightly from actual prices, suggesting prices to be quite accurately recalled. The average recalled price of laundry detergent was the closest to the actual price, which differed only 6 cents. However, the mean absolute deviation percentages (MADP) illustrate a different image: prices recalled for laundry detergent are actually relatively the most far off, with the relative worst MADP score for laundry detergent. The average absolute deviation percentage of all observations is 21.4%, which is in line with the MADP values of recent studies (Eisenhauer and Principe 2009). When looking at the percentage of correctly recalled prices, product coffee is recalled the lowest amount of times, followed by laundry detergent. Prices for the product beer are correctly recalled the highest amount of times, which is in line with the lowest average deviation percentage for this product. Judging from these statistics, it seems that price knowledge is relatively the most accurate for the product beer. The total share of respondents that correctly recalled prices is 6.5%, which mediates the found percentages of consumers that correctly recalled prices of Vanhuele and Drèze (2002) of 2.1% and Jensen and Grunert (2014) of 7.6%, the latter being prices recalled before shopping.

Table 6: descriptive statistics of price recall question values.

To get an indication of the differences in absolute price knowledge per product and per brand of the actual measures that will be used to assess price knowledge, a visual representation of the absolute deviation percentages for deal spotting and price recognition per brand can be found in figure 6. This

Product Average recalled price Actual Price MADP Correct price recalled

Coffee €4.35 €4.94 24.7% 4.6%

Laundry Detergent €6.17 €6.23 25.1% 5.8%

(38)

figure indicates the least accurate absolute price knowledge for the product coffee, a slightly more accurate absolute price knowledge for the product laundry detergent and a clearly more accurate absolute price knowledge can be concluded for the product beer, the latter being partly caused by a lower range of prices used for price recognition. These results are consistent with earlier conclusions with regards to the price recall measure. Additionally, it can be concluded that respondents reveal the relatively highest consumer knowledge of the product beer, since for all measures of product and price knowledge, the most accurate scores can be concluded for this product.

Figure 6: Average absolute price estimation error in percentages per product per brand for measures price recognition and deal spotting.

A second way of looking at the price knowledge variable is to consider underestimation and overestimation, looking at the actual deal spotting and price recognition scores. The average price knowledge accuracy including under- and overestimation per brand is represented in figure 7. From this figure, it can firstly be noted that for all brands except one, the prices are on average underestimated. Frequent price promotions, which lead to the lowering of expected prices (Lowe, Yee and Yeow 2014), might explain the average underestimation of prices. The one exception, prices for brand Kanis & Gunnink are overestimated, could be explained by the difference in the normal price between Kanis & Gunnink and the other two coffee brands, ranging between 25 and 28% below Douwe Egberts and Van Nelle’s normal prices. This dissimilarity in normal price could likewise be the explanation for the difference in average underestimation between Ariel and the other two laundry detergent brands, since the normal price of Ariel ranges between 10 and 17% above the normal price

11,2% 11,9% 13,6% 11,1% 11,0% 10,0% 7,1% 7,4% 6,8% 5,8% 5,7% 5,7% 5,6% 5,4% 5,2% 3,7% 4,1% 3,7% 0,0% 2,0% 4,0% 6,0% 8,0% 10,0% 12,0% 14,0% 16,0%

Price estimation error in absolute

percentages

(39)

of brands Omo and Robijn. The third observation that could be made is that the prices of product category beer seem to be underestimated with a constant percentage, the slightly higher price (around 3%) of brand Hertog Jan compared to Heineken and Grolsch seems to be acquainted with by the public. To simplify model estimation, the price knowledge measures per brand were combined to form an average price recognition and deal spotting value per product. To make it possible to assess the different effects of price knowledge on the WTP of respondents that underestimated and respondents that overestimated the normal price, both the price recognition and deal spotting variable were split into two separate variables: one variable containing the values for respondents that underestimated, one containing the values for respondents that overestimated, the empty spots filled with zeros. Chapter 4 compares the predictive ability of the absolute price knowledge variables with the actual price knowledge variables, which led to the adoption of one of the two sets of variables in the linear model.

Figure 7: Average price knowledge accuracy including under- and overestimation in percentages per product per brand for measures price recognition and deal spotting. Average actual prices per brand

are illustrated below the graph. 4.1.4 Store Price Image

(40)

reveal a clear distinction between the three different stores, as illustrated in table 7. Supermarket chain Plus has the highest store price image score, which represents the least favourable price image. Consumers reveal the most favourable price image for supermarket Jumbo, with a slightly more expensive price image indicated for Albert Heijn.

An association between the choice of supermarket and store price image seems to be present, with a significantly negative correlation between the choice for Albert Heijn as favourite supermarket and the store price image of Albert Heijn, a significantly positive correlation between choice for Jumbo or Plus as favourite supermarket and store price image of Albert Heijn, a significant negative correlation between choice for Jumbo as favourite supermarket and store price image of Jumbo, a significant positive correlation between choice for Albert Heijn as favourite supermarket and store price image of Jumbo, and a significant negative relationship between choice for Plus as favourite supermarket and store price image of Plus, indicating an association between favourite supermarket and a lower, more favourable price image which vice versa holds partly. Table 7 furthermore illustrates this, with a substantially more favourable price image for the preferred store.

Table 7: average store price image (higher values mean less favourable price image) per store, total and if the store is preferred plus relative preferences per store.

Since the SPI of the preferred store is of interest in this research, relative SPI scores were calculated next, by dividing the SPI of the preferred store on the average SPI across all three supermarkets. This way the SPI data is normalized, which improves the interpretability of the data. The new variable tested the effect of an increase in the relative SPI of the preferred store, compared to the average SPI of all three stores.

4.2 O

BTAINING

W

ILLINGNESS

T

O

P

AY

V

ALUES

In this section, the choice-based conjoint data is analysed to obtain WTP values for the preferred supermarket for each consumer. The first step in obtaining WTP values per product for each individual was dividing each dataset in an optimal number of segments, by way of a latent class analysis. These classes are latent in the way that each consumer belongs to a class with a particular probability (Eggers et al. 2015). To classify each respondent as precisely as possible, covariates by way of demographic variables were added to the model. To assure a better interpretability of the different segments, each demographic variable was transformed to it consisting of three levels, for instance, the income variable

Store Price Image Total If preferred Relative Preferences

Albert Heijn 3.96 3.43 50.2%

Jumbo 3.78 3.07 39.1%

(41)

was transformed to below average (<€20.000), average (€20.000-40.000) and above average (>€40.000) income.

The optimal number of segments was determined by information criteria and entropy values (Eggers et al. 2015). With regards to information criteria, the optimal number of segments would be 7 for coffee, 8 for laundry detergent and 8 for beer. However, this number of segments illustrated illogical parameter estimates for the price variable and R² values. When dividing the datasets of the products coffee and laundry detergent in 4 classes, the BIC and CAIC information criteria do not vary too much from the optimal number, and the classification error is amongst the optimal values. For the product beer the data has to be divided by 5 classes, since the information criteria, as well as the classification error, improve significantly compared to a segmentation in 4 classes. Illogical values were furthermore avoided by these smaller number of segments for each product, leading to the adoption of this number of classes for the utility models. The information criteria, R² and classification error per product for the selected number of classes are illustrated in table 8.

Table 8: information criteria, R² and classification error for the selected number of classes per product.

Next, WTP values per segment per supermarket were calculated. The absolute WTP for the product in the preferred supermarket compared to the no-choice option per segment was calculated by determining the price for which the average of the utilities per brand for the preferred supermarket equals the utility of the no-choice option, as used by Gensler et al. (2012). In other words, the WTP for a product is calculated as the price for which a consumer equally values purchasing and not purchasing the product (Gensler et al. 2012). Size, utilities per attribute, WTP values and significant characteristics per segment are illustrated in table 9 for product coffee. The differences in size, utilities, WTP values and significant characteristics for the products laundry detergent and beer can be found in appendix 8.

Coffee (4 classes) Detergent (4 classes) Beer (5 classes)

BIC 2085.9 2304.2 2430.1

CAIC 2139.9 2358.2 2500.1

R² .6866 .6614 .6812

(42)

Note: a p-value <.01, b p-value <.05, c p-value <.10

Table 9: Size, utilities, willingness to pay and characteristics per segment derived from the latent class analysis for the product coffee.

The first segment, consisting of mostly males with an above average size household, has a strong preference for brand Douwe Egberts and supermarket Albert Heijn. Their WTP for the product coffee is amongst the highest values. Segment 2, which consists of mostly females in a slightly below average sized household, prefers brand Kanis & Gunnink, and is pretty indifferent with regards to supermarket choice. Their WTP for the three supermarkets is quite alike. The third segment, consisting of mostly females with a small-sized household strongly prefers Douwe Egberts, while strongly disliking brand Van Nelle, and is quite indifferent with regards to supermarket choice. This segment revealed the highest value of the no-choice option together with a strong price sensitivity, resulting in the lowest WTP for the product coffee. The fourth segment, consisting of mostly females with fairly large sized households, is quite indifferent with regards to brands, but strongly prefers supermarket chain Albert Heijn. When looking at the segments for the product laundry detergent, the WTP values across supermarkets do not differ much except for segment 3. The second segment displays the highest average WTP values, the third segment the lowest. Segment 4 revealed the highest preference

Coffee Segment 1 Segment 2 Segment 3 Segment 4 Wald

Size 66 57 33 15

Utilities

Brand: Douwe Egberts 3.50 0.12 2.56 -0.08 125.43a

Brand: Kanis & Gunnink -1.16 0.70 0.46 -0.02

Brand: Van Nelle -2.33 -0.82 -3.02 0.10

Chain: Albert Heijn 1.24 0.08 0.35 2.80 45.40a

Referenties

GERELATEERDE DOCUMENTEN

Therefore, this study on a support group for childless women and men in Ghana contributes to anthropological insights into the activities of support groups and counseling

We have presented two ways in which to achieve coordination by design: concurrent decomposition, in which all agents receive tasks with an additional set of constraints prior

(1997) report that contractionary monetary policy accounts for the decline in aggregate economic activity following an oil price increase. If prices are sticky downward, an oil

These results show that in the case of the product category butter, an increase of a SKU’s number of facings, which is higher priced than the consumer’s reference price of

This subchapter has been included in order to analyse whether, and to what extent, differences in mean scores of the 4 P’s Price, Product, Promotion, Place and the main

• The higher / lower (as compared to the mean attitude) a respondent’s indicated attitude, the higher / lower their respective concession on either price or quantity. • The higher

! Price and Quantity Framing did not reduce feelings regarding difficulty. and wrongness during the

Furthermore, the higher consumers perceive the Aldi as hedonic (high quality products, high prices, high service level, large assortment) the less a consumer wants to pay a