Optimal delivery conditions for an online retailer

(1)

Master Thesis, Thomas Wiegerinck

University of Groningen

28-06-13

(2)

Optimal delivery conditions for an online retailer

by

T.H.A. Wiegerinck

University of Groningen, Faculty of Economics and Business

Master Thesis

Theme: Choice based conjoint analysis Msc. Marketing Intelligence

28 juni 2013

Address Visserstraat 52, 9712 CX Groningen

Phone +31611517291

Email thomaswiegerinck_23@hotmail.com Student number 1793721

First supervisor Dr. M.C. (Marielle) Non

(3)

MANAGEMENT SUMMARY

This research focuses on the delivery process of an online retailer aiming to find out what condition of the delivery process for online retailers is best valued by the customer and if segments could be distinguished. The condition of the composited delivery process in this research differ in product availability, recommendations of other consumers regarding order fulfillment, delivery costs, delivery time and price of the product. The method that has been used in this study is a choice based conjoint analysis in order to estimate preferences for delivery conditions, because if online retailers better understand their customer’s needs, they can present better services regarding the delivery process.

In a survey 207 respondents were asked to choose under which delivery condition they would purchase a T-shirt. With this data two models have been estimated, one on aggregate level the second one with different classes in order to find differences in preferences among customers. The aggregate model shows that the most important attribute is ‘delivery costs’ followed by ‘price’. Meaning that deciding upon which retailer to purchase from monetary value is the decisive factor. Recommendation about the delivery process is being perceived as least important attribute of the delivery process.

A latent class analysis has been used in order to find different segments. The analysis revealed different preferences for consumer segments regarding the delivery conditions and in total four different segments were distinguished. One segment being ‘cost conscious’, with the delivery costs as most the important attribute of the delivery process. Another segment with ‘bargain hunters’ focusing on price. A segment called ‘rationalists’ considering all the different conditions somewhat equal with a slightly more emphasize on recommendations about the delivery process. Last segment that was found are those who are ‘time focused’, meaning that product availability and delivery time were appreciated the most. These found segments indicate that online retailers should therefore carefully consider their customers when deciding upon how a delivery process should be designed in order to gain a competitive advantage.

(4)

PREFACE

After my trip of five months through South-east Asia I decided to get my Msc degree in marketing. Which degree in marketing was not clear yet, until it became clear during the study that Marketing Intelligence was the direction of interest for me. I am very glad that I have chosen for the degree of Marketing Intelligence, since I enjoyed the Marketing Intelligence courses very much. And now this master thesis represents the final phase of my study. I used my gathered knowledge to write my master thesis and in front of you lies the end result.

I would like to thank my supervisor dr. M.C. (Marielle) Non for her patience, useful insights and guidance during the writing of this thesis. She was always very fast with giving feedback, which I appreciated very much, so thank you for that. Furthermore I want to thank my thesis group for providing help when it was needed. I really enjoyed the time I spent with them during the writing of my thesis.

The time has come to say goodbye to my student life, which I really enjoyed together with my friends. The next phase of my life is just about to start and it is time to apply for a job. I am ready and very eager to work in the competitive marketing environment.

Thomas Wiegerinck

(5)

1. INTRODUCTION

The purchase process is rapidly changing these years. Consumers and retailers are shifting from the offline world to the online world. Research at the interface of online retailing is growing in popularity. Not strange given the fact that in 2001 the B2C e-commerce market was estimated at 31 billion dollar in the USA and this market has grown to more than 150 billion dollar in 2010 and is still growing (Griffis et al. 2012). There is a shift in competition among retailers, a shift from the traditional brick and mortar stores to online retailing. This has consequences for the way of doing business. Many studies showed that the online shopping environment and behavior of customers is fundamentally different from that of the conventional environment (Evanschitzky et al. 2004, Liu et al. 2008). Therefore retailers should seek again for the best possible experience to satisfy their consumers online, where consumer satisfaction can be defined as an attitude formed through a mental comparison of the service and product quality that a consumer expects to receive from an exchange with the level of quality the consumer perceives after actually having received the service/product (Kim et al. 2009). The amount of competition is starting to increase among online retailers. In this competitive environment, coupled with the increase of consumers and different consumer characteristics, the understanding of key factors that drive consumers in online shopping has become increasingly important (Ha and Stoel 2012).

(8)

2 | P a g e

delivery process are separated temporally. This has implications for the delivery time and therefore product availability is part of the delivery process. However none of these studies takes all the different aspects into account and combines them at different levels to create a trade-off with regard to delivery aspects. It is commonly known that consumers have different preferences regarding services and this has been investigated many times (Ostrom and Iacobucci 1995). Therefore this research focuses on the different conditions of the delivery process together, because if online retailers better understand their customer’s needs, they can present better services regarding the delivery process, in order to strengthen their competitive advantage and satisfy their customers. Next to these delivery conditions price will be taken in to account as well, because price and delivery costs are closely related to each other. Online retailers offer sometimes partitioned prices, consisting of the net product price and delivery costs. Above all equalization prices can be calculated when a price is known and is of great value as will be discussed later on in this research. It is expected that preferences regarding the delivery process can be distinguished as well. This paper addresses attention to these differences in preferences regarding the delivery process for consumers. It can be reasoned that some people cannot wait until the product arrives, whereas others value the best price possible and do not care about the delivery time. What does this mean for the strategy of online retailers and can customer segments be distinguished and served? More specific, what is most important for consumers regarding the delivery process and what are valuable attributes were online retailers should pay attention to.

This research aims to find out which delivery conditions offer the highest level of utility to the customer. Therefore this paper investigates which of the common conditions regarding the delivery process generates the highest utility for the consumer. The conditions differ in product availability, recommendations of other consumers regarding order fulfillment, delivery costs, delivery time and price of the product. This leads to the central research question of this paper.

“What condition of the delivery process for online retailers is best valued by the customer and can customer segments with different preferences for delivery conditions be

(9)

3 | P a g e

The theoretical and practical relevance are now discussed shortly. First of all, this research contributes to the current knowledge about strategies for online retailers and especially for the knowledge about the delivery process. Research concerning online retailers is growing rapidly. However extensive research about consumer trade-offs regarding the delivery process of online retailers seems lacking in the academic literature. This topic, to the author’s knowledge, has not been previously explored in the extant academic literature. To address the lack of research on the delivery process of online retailing and the consumer trade-offs regarding preferences, this study focuses on different attributes of the delivery process and are all judged together. Respondents are asked to evaluate different delivery processes to estimate preferences among consumers. Hence, a basis is created for future research, which can develop strategies to satisfy different needs of consumers regarding the delivery process.

Finding the different trade-offs for consumers regarding the delivery process results in the practical relevance of this research. It is of great value for marketers to know how the delivery process influences the purchase decision of the consumer without taking the products into account. This knowledge can be used to increase satisfaction of consumers and to gain a competitive advantage to win the severe competition online. This can result in higher total sales for the online retailer and furthermore marketing efforts are more accountable, which is becoming more and more important in the marketing field (Verhoef and Leeflang 2009).

(10)

4 | P a g e

2. LITERATURE REVIEW

In this chapter the different attributes regarding the delivery process will be described, the attributes taken into account are product availability (i), delivery time (ii), delivery costs (iii), price of the product (iv) and recommendations regarding the delivery process (v). Furthermore some moderators will be described and discussed.

2.1 Product availability

Traditional brick-and mortar stores can differentiate themselves via store location, store atmosphere, interaction with staff in store, merchandising attractiveness or staff friendliness. Online retailers have to put the focus on other variables for achieving differentiation. A variable that becomes a primary factor in online retailing is product availability (Taylor et al. 2004). A key feature of an online retailer is the flexibility in managing the inventories. A brick-and-mortar store has to put some inventory on its shelves before any products can be sold, the contrary is true for an online retailer. A product can be sold even if it is not in stock, because the purchase decision and the delivery process are separated temporally (Zhaoa and Cao 2004). It means that the customers only view images of the products that are offered by the online retailer rather than the product in physical appearance. It also means that customers can buy the product even if it is not in stock, but this has consequences for the delivery time.

(11)

5 | P a g e

An online retailer should never forget the long time relationships with the customer. In order to reduce dissatisfaction of customers, while shopping online, online retailers can provide extra information. For example an indication when an out of stock situation is occurred or when the product will be available again (Breugelmans et al. 2006). Online shoppers attach much value to the in-stock information, it is being considered as an important service element with regard to online purchases (Kim and Stoel 2005). Furthermore it is critical for online retailers to manage shopper expectations of product availability so that online shoppers are well informed of product availability at an early shopping stage (Kim and Stoel 2005).

The psychological reactance to product scarcity from a marketing perspective has been studied by Clee and Wicklund (1980). They stated that when a product is scarce the product becomes more attractive, it encourages consumer’s choice to buy, because of uncertainty of not obtaining the product. This is an important aspect when it comes to online retailing. In e-commerce it is very common to show the number of available items on a website when a consumer is searching for a product where he or she is interested in (Kurata and Bonifield 2007). An online retailer can for example show the number of available products to encourage the consumer to buy, because of fear of not getting the product, creating a sold out situation aversion. Important for an online retailer is to find out if showing product availability affects consumer’s online purchase decisions. Inventory shortages can have adverse consequences on both short term and long term customer behaviors (Jing and Lewis 2011). For the short term behavior of the consumers Jing and Lewis (2001) found that lower stock out rates are associated with additional ordering and consumers tend to increase buying in categories in which a stock out occurred. The long term behavior reflects that cumulative stock outs have a negative effect on long term retention.

(12)

6 | P a g e

at the online retailer and wait until the product is available again (iv) do not buy the product, which means a lost sale as well, (v) or buy the product and accept longer delivery time.

We are particularly interested in the ‘go to another store response’ of a consumer. Since the goal of this research is to estimate the preference structure of consumers regarding aspects of the delivery process in comparison between three available online retailers. Since consumers are likely to postpone, go to another shop, do not buy or try to find a substitution for the product the following hypothesis H1a is expected. It is expected that the difference in utility between an out-of-stock situation and limited stock available is larger than the difference between limited and full stock, therefore H1b is expected.

H1a: Product availability has a positive effect on the level of utility for the customers.

H1b : The difference in utility between an out-of-stock situation and a situation with limited

stock is larger than the difference between limited stock and full stock. 2.2 Delivery time

(13)

7 | P a g e

al. 2000). Therefore if an ordered product does not arrive at the agreed time, dissatisfaction about the online retailer may arise.

Online retailers must understand what drives consumer satisfaction. When a consumer expects the product being on time, and it is not, satisfaction drops dramatically which result in lower repurchase intention (Posselt and Gerstner 2005). Therefore it is of great importance to meet expectations of the consumer and deliver as agreeed. On-time delivery significantly influences the overall rating of a website (Posselt and Gerstner 2005). The aim of this research is to find out what drives consumers to choose for a website to buy from, but nevertheless repurchase intentions are important for getting market share and survive the severe competition among online retailers. Furthermore on-time delivery is being considered by consumers as a satisfier and a poor delivery and logistic performance is being viewed as unfavorable by consumers who order a product online (Gwo-Guang Lee and Hsiu-Fen Lin 2005).

Whether and how people incorporate the future in their decision making about consumption is of great importance for people ordering online. Consumers are faced with the threat of an unexpected delay in receiving the product that they ordered. Chen et al. (2005) examined cross cultural variations in how people are discounting time. It seems that Eastern cultures are more patient and can wait for consumption whereas Western people value immediate consumption relatively more. This is an important finding which we can translate to the delivery time of an ordered product. Since Western people, where we focus at in this investigation, are more eager for immediate consumption, delivery time should be shortest possible.

(14)

8 | P a g e

deliveries resulted in more referrals, which implies that faster delivery is preferred. Therefore the following hypothesis is expected

H2: The delivery time of a product has a negative effect on the level of utility for the customer

2.3 Delivery costs

Consumers perceive the internet as transparent to compare products among different sites and web-shops. However the truth is that despite the fact price comparison should be a simple process, prices are still subject to a great deal of dispersion, because of hidden delivery costs (Ratchford 2009). This brings us to the problem of delivery costs at e-commerce. Because what do consumers value when it comes to delivery costs.

Retailers are splitting the gross product price into partitioned prices, consisting of net product price and delivery costs. The reason why they act on this pricing strategy is that partitioned prices exert a positive influence on the intention to buy for consumers (Xia and Monroe 2004). For an online retailer this means that when offering a partitioned price the likelihood to buy increases. Morwitz et al. (1998) found in their study the same positive influence on the consumer’s intention to buy. They concluded that offline retailers are aware of the fact that the price setting in a partitioned way lowers total perceived costs. Although the retailers are operating in an offline environment, the principle is the same for operating online. Consumer’s biased perceptions of partitioned prices create opportunities for the retailers operating online to win the price comparison based on the net product price but eventually charge for higher delivery costs.

On the other hand free delivery offers help to reduce the drop out in the check-out stage, which means that more consumers are refrained from dropping out in the check-out stage (Bertini and Wathieu 2008). But when offering free deliveries, someone has to pay. It is more like a trade-off between lower net price but higher delivery costs or higher net price and no delivery costs.

(15)

9 | P a g e

is available, preferences for pricing formats do emerge. Consumer who are skeptic about delivery costs then prefer the offer in a singular price and consumers who are less skeptic about delivery costs prefer the price in an unbundled format. Furthermore the authors argue that when a skeptic delivery costs consumer is offered a price in an unbundled format, skeptics about delivery costs having the uncomfortable feeling that they are paying two profit generating prices instead of one. They may feel that the seller’s attempt to claim otherwise is “sneaky and unfair”. The none skeptics think it is “fair” to pass the delivery costs to the consumers and thus an unbundled price is less painful to pay.

Shampanier et al. (2007) conducted several experiments to investigate the effect of (zero price) free offers. The results showed that the free (zero price) options are evaluated more positively than the other options which were not free. A price of zero means that it adds to the benefits of consumers, because humans are intrinsically afraid of losses. So a price of zero does not only decrease perceived costs. The results of Shampanier et al. (2007) provide support for the affective evaluation idea, this means that a free offer (price zero) has a stronger positive effect than would be predicted by a normal cost benefit analysis. So the difference between a price of zero and one cent is perceived much higher than a price difference between one cent and two cent. And so consumers are perceiving a more positive affect when they are being offered a free offer, which leads to a higher demand for the (zero price) offer (Finucane et al. 2000). Ariely (2009) found this same theory of reasoning, in his study people were offered a choice of a fancy Lindt truffle for 15 cents and a Hershey’s kiss (chocolate) for a penny. This resulted in a large majority choosing the fancy Lindt truffle (73%). But when offering the same chocolates for one penny less each – Lindt truffle (14 cents) and Hershey’s Kiss (free) - only 31% selected the Lindt truffle. Therefore the word free Ariely (2009) discovered is a powerful tool that can even turn customers away from a better deal and toward the free one, stating in this research that the Lindt truffle is a far better deal than the Herhsey’s kiss.

(16)

10 | P a g e H3a: Delivery costs for a product have a negative effect on the level of utility for the

customers

H3b: There is a stronger negative effect between free delivery costs and €1,00 than between €1,00 and €4,95 delivery costs

2.4 Price

Price has always been an important stimuli for buying products. Whether a consumer is buying online or offline, price is one of the most important decision influencers in the decision making process for a consumer when buying a product (Kuan-Pin Chiang and Dholakia 2003). A price can be defined as: “the consumer's perceptual representation or subjective perception of the objective price of the product”(Kuan-Pin Chiang and Dholakia 2003). Dutch consumers are basing their online purchase decisions mostly on price, research of Intomart GfK (2013) showed that 89% of Dutch consumers based their decision on the decisive factor price. When a price is not as low as expected, buyers looking for low prices will be disappointed, resulting in a negative attitude toward the online retailer and will result in a lower utility for the consumer (Cao and Zhao 2004).

But how do consumers evaluate prices? As search costs increases, online retailers will raise their prices to take advantage of consumers seeking to reduce their effort of finding lower prices. Those consumers who do not extensively search and put time in searching will end up paying higher prices (Baylis and Perloff 2002). Consumers vary in the ability and in their motivation to search for best prices (Johnson et al. 2004). In the research of Johnson et al. (2004) the authors found that people visit few stores despite the fact that consumers can easily navigate to other stores, furthermore the search behavior varied by product category and the level of activity, but showed no increase with experience. Also Luo and Chung (2010) found that online retailers charged lower prices when the number of retailers offering the same product was high. This implies that consumers are eager to find the lowest possible price.

(17)

11 | P a g e

monopoly or an advantage any one retailer might have over another because of its proximity to the consumers.

The internet has made the process of finding and comparing prices of products significantly easier for shoppers (DiRusso et al. 2011). The difference between the online and offline marketplaces is the comparison of prices. In the offline world, prices cannot so easily be estimated and compared by the consumers. Transparency regarding price in the traditional brick-and-mortar stores is harder to obtain for a consumer than when a consumer is buying online. A barrier to finding the lowest prices for a product in the brick-and-mortar store has always been the lack of information and the cost of acquiring this information (Reibstein 2002). It means walking through the city for hours and finally decide upon which product is cheapest to obtain. The offline search costs for the customers are very high. However with online shopping prices can be viewed and compared in different stores by a few clicks more. The process of searching for information online has even more accelerated with the availability of electronic agents that automate the search process. Price dispersion can be described as the variation in price for the same product and it can be defined as: “the distribution of prices of an item with the same measured characteristics across sellers, as indicated by measures such as range and standard deviation of prices” (Pan et al. 2004). It was expected that internet would play a major role in the disappearance of price dispersion, because of the fact that prices could be easily compared with competitors (Bakos 1997). In electronic markets prices can be found and compared for the same product much easier then while searching in offline markets. Evidence however shows that there is still price dispersion in online retailing, despite low search costs and consumers’ access to information (Lindsey-Mullikin and Grewal 2006), which is a contradictory finding regarding the study of Bakos (1997). But why is there a price dispersion when the seller knows there is information available and easily accessible for every consumer? Other variables can influence the price in online retailing ,like privacy, the customer service provided, delivery and product information and therefore perpetuating price dispersion (Brynjolfsson and Smith 2000; Degeratu et al. 2000).

(18)

12 | P a g e

advantage besides other elements of differentiation. It is expected that a higher price will result in a lower utility for customers and this results in the following hypothesis.

H4: The price of a product has a negative effect on the level of utility for the customers.

2.5 Recommendation

Trust has been defined by Rousseau et al. (1998) as: “a psychological state comprising the intention to accept vulnerability based on positive expectations of the intentions or behaviors of another.” Specifically, the trust a consumer has in a party is the willingness to accept vulnerability but with a certain state of confidence and expectation that it can rely on the other party (Morgan and Hunt 1994). The study of Morgan and Hunt (1994) focuses on relationship-marketing, which is often studied in the marketing literature. The trust that often has been studied is based on buyer-seller relationships in the offline world. Studies have shown that trust in sales agents evolves over time and is subject to the party’s honesty and reliability (Doney and Cannon 1997). But in fact the principle is the same in the online world, more specifically for an online retailer, where the party’s (online retailer) honesty, fulfills the expected agreement towards the buyer. This means that the consumer gets what was agreed on, more specifically, the product that was ordered by the consumer. Reliability in the online retailer can be a form of timely delivery, so that the product arrives on the expected time. In order to enhance consumer satisfaction and more important purchase intentions, online retailers should focus on improving the reliability dimensions, for example the capability of delivering products as promised. But also up-to-date and accurate information and strengthening the security of online transactions (Gwo-Guang Lee and Hsiu-Fen Lin 2005). Given the importance of trust in transaction relationships it is essential to understand online trust.

(19)

13 | P a g e

order fulfillment track record of that website. Wolfinbarger and Gilly (2003) indicated that order fulfillment is an important attribute contributing to the quality of an online retailer and therefore an indication of trust and as said this order-fulfillment can be translated into a form of a rating of a website, either by third parties or consumers. When consumers find reviews about sellers, they are able to better interpret experiences about previous transactions, therefore they can better assess the level of trust they have in a particular web shop (Petrescu 2011). These reviews reflect word of mouth of previous customers.

Word of mouth communication which can be defined as: “the exchange of information about goods and services among consumers” has long been recognized as a valued an influential source consumer information (Schindler and Bickart 2012). The internet increased this WOM communication, particularly in the form of consumer reviews of an online retailer. Furthermore research has suggested that electronic WOM such as consumer forums generates greater empathy, credibility and relevance then does information generated by the organization itself (Bickart and Schindler 2001). The online reviews consumers provide to each other can have different characteristics with respect to the wording of an online review. The review can be divided in content and style, but that is not taken into account in this research. The goal in this research is to find out to what extent the online review about the delivery process transferred into a rating system is affecting the level of utility of a customer.

(20)

14 | P a g e

It is expected that the higher the level of the recommendation the higher the utility is for the consumer. This result in the following hypothesis.

H₅: The provided level of recommendation of the delivery process has a positive effect on the level of utility of a customer

2.6 Moderators

In the next section three moderators will be described based on literature. The variables impatience, rationality and attaching value to the opinion of others are those that possibly affect the strength of the relation between the utility and some attributes. The moderating effect these variables may have will be briefly described.

2.6.1 Impatient customers

(21)

15 | P a g e H6a : The positive effect on the level of utility for product availability is stronger for

impatient people than for people who are patient.

H6b: The negative effect on the level of utility for delivery time is stronger for impatient

people than for people who are patient. 2.6.2 Rationality

Three very different views of the way individuals make choices underlie contemporary research in choice modeling namely the economic view, the behavioral and psychological view and the statistical view (Adamowicz et al. 2008). Drawing on aspects of both psychology and economics the operating assumption of behavioral economics is that cognitive biases often prevent people from making rational decisions. The way people arrive at decisions to buy a product at a particular price is thought of school for behavioral economics. The behavioral and psychological view argues that if preferences of consumers even exist, they are lumpy and not accurate. Preferences are merely constructed at the time of choice and are based on situational/contextual factors. Preferences for choice alternatives result from preference over much more proximal sources of satisfaction (Schwarz 1999). Many experiments have shown that people are incapable of making good decisions (Ariely 2009).

The economic view of decision making and thus rational decision making takes a perspective that people make choices in ways that are consistent with random utility maximization. This means that consumers may be assumed to have well developed preferences, defined narrowly over product attributes and the choose alternatives with attribute levels that offer the best trade-off (Adamowicz et al. 2008). Rational people follow a logical way of thinking. Logical thinking means in the scope of this research that delivery costs and price are equally important for rational people, since it is reflected by a monetary value and it can be compared with each other. Therefore expectation is that for rational people the attributes ‘delivery costs’ and ‘price’ are equally important. This theory results in the following hypothesis:

H7: Rational people assign equal importance to ‘price’ and ‘delivery costs’.

2.6.3 Attaching value to opinion of others

(22)

16 | P a g e

previous section. Definition of Sweeney et al. (2008) is more specific and they defined WOM as: “a process of personal influence, in which interpersonal communications between a sender and a receiver can change the receiver’s behavior or attitudes” (Sweeney et al. 2008). Again it is about the exchange of information, but more focusing on the personal influence and outcomes, meaning that only the generation of positive WOM is not sufficient for it to be an effective source of communication. WOM concerns sender and receiver, whereby the personal factors of the receiver play an important role as well and the possible change of the receiver’s behavior or attitude.

Researchers have found that consumers are more likely to seek other people’s opinions before purchasing when they have less experience and a stronger involvement in the purchase of the product category (Gilly et al. 1998). We can apply this line of reasoning to the delivery process of an online retailer. Furthermore people who perceive more risk in a purchase situation tend to seek information through WOM (in this case the level of recommendation of the delivery process of previous customers) more actively than those who perceive a lower risk (Bansal and Voyer 2000). Hence people who are less experienced and are afraid of not getting the product delivered attach more value to the recommendations of the delivery process than people who perceive a lower risk. Apart from experience level and trust issues, people can attach value to the opinion of others for another reason. An easy way of evaluating the possible options can be: “because others did it and therefore I will do it as well”. This is a consumer motive for online opinion seeking accordingly to Goldsmith and Horowitz (2006), and so these people attach value to the opinion of others in another way. In case of ordering a product at an online retailer A or B, the consumer proceeds to purchase at the one with a high recommendation rating. This theory results in the following hypotheses.

H8a: The positive effect on the level of utility for recommendation is larger for people who attach value to the opinion of others

(23)

17 | P a g e

3. CONCEPTUAL MODEL

The expected effects found in the literature are given in the conceptual model as visible in figure 1. Some moderators will be shortly explained in more detail.. For hypothesis 1b the

higher the level of SKU available the less stronger the effect is on utility for product availability, meaning a negative relationship. From the literature it was found that there exist a stronger negative effect when there is a small difference between a free delivery and delivery costs than for two options of having both delivery costs, therefore hypothesis 3b is

marked as positive. For hypothesis 6a and 6b, the more impatience means the stronger the

effect of product availability and delivery time, therefore it is marked as positive. Hypothesis 7 could not be included in the conceptual model. People who value the opinion of others and attach importance to that are expected to have a stronger effect on utility for attribute recommendation, this means a positive relationship for hypothesis 8a. Also little experience

in online shopping will result in a stronger effect on utility and therefore hypothesis 8b has a

positive relationship. All the other relationships are self-explanatory.

(24)

18 | P a g e

4. METHODOLOGY

This chapter will describe which methods will be used in order to find empirical evidence to support the given hypotheses. First there is described which conjoint analysis will be used, after that the attributes and attribute-levels will be presented followed by a model specification. Subsequently the study design is presented and there will be explained why a factor analysis is used. This chapter will end with explaining the last applied method, an OLS regression.

4.1 Conjoint analysis

Conjoint analysis is a multivariate statistical technique used in market research to determine how people value different attributes that make up an individual product or service. In this research five different delivery conditions (attributes) and three levels of delivery conditions (attribute levels) are evaluated. The objective of a conjoint analysis is to determine what combination of limited number of attributes is most influential on respondent’s choice or decision making. Multivariate analysis is a technique that analyses simultaneously multiple measurements on individuals under investigation (Hair, Black, Babin & Anderson, 2009). Conjoint analysis assumes that any set of objects or concepts is evaluated as a bundle of attributes. Having determined the contribution of each factor to the consumer’s overall evaluation, the researcher is able to, define the object or concept with the optimum combination (i), show the relative contributions of each attribute and each level of the overall evaluation of the object (ii), use estimates of consumer judgments to predict preferences among objects with differing sets of features (iii), isolate groups of potential customers who place differing importance on the features to define high and low potential segments (iv) and last identify marketing opportunities by exploring the market potential for feature combinations not currently available (v) (Hair et al, 2009). Especially (iv) is useful for companies, because of the marketing purpose to target potential segments with favorable conditions if their preferences are known. Therefore later on in section 4.4 a model specification is given with segments included, because that is also in the interest of this research to find customer segments with different preferences for delivery conditions.

4.2 Choice based conjoint analysis

(25)

19 | P a g e

respondent’s actual decision process (Moore 2004). The traditional rating based models and choice based conjoint models are different in predicting choice probabilities, because traditional models convert preference ratings to choices through rules like maximum utility that have rather arbitrary assumptions and choice based models predict choice probabilities directly (Moore 2004). Since the object of a conjoint study is the prediction of choice or market share, it seems more logical to derive the parameters of the choice process of interest (delivery conditions in this study) from choice data rather than from a (perhaps) parallel judgment of evaluation task (Louviere and Woodworth 1983). The choice based conjoint is more representative for this study of selecting delivery conditions from a set of different online retailers. Due to the more complicated task, the number of attributes included is more limited compared to the traditional conjoint analysis. In this research five attributes are included with three levels and is suitable for a choice based conjoint analysis. In a choice based conjoint a “none option” can be included as well for calculating willingness to pay. However in this research the “none option” is not included, because this research investigated the hypothetical situation where the respondent already has decided to buy a product but has to choose between different delivery conditions.

4.3 Attributes and attribute-levels

(26)

20 | P a g e

Table 1: Attributes and attribute-levels

The levels are based on what is most common in the Netherlands. The product availability has three levels, more than 10 SKU available, between 0-10 SKU available or not available. The delivery time varies between one day, three days and five days. Delivery costs are respectively €0,00, €1,00 or €4,95. The next attribute-levels are the prices of the product, which are respectively €29,95, €32,95 and €34,95 The recommendation is expressed in a rating system based on stars, the rating system consist of a maximum score of five stars and a minimum score of zero, where scoring five stars means an excellent delivery process that is reliable and conform agreement and zero stars means that the delivery process is evaluated badly. The levels included here are five stars, four stars and three stars. Not less than three, because that is not representative. It can be assumed that when an online retailer is reviewed with one or two stars the online retailer would not survive the severe competition.

4.4 Model

Now that the attributes and attribute-levels are known a model can be specified. Where, U= Utility j= segment 1,…,n X₁ = product availability X₂ = Delivery time X₃ = Delivery costs X₄ = Price X₅ = Recommendation

Attribute Level 1 Level 2 Level 3

Product availability > 10 SKU

available

Between 0-10 SKU available

Not available, will be delivered when available again according provided delivery time

Delivery time 1 day 3 days 5 days

Delivery costs €0,00 €1,00 €4,95

Price of product €29,95 €32,95 €34,95

(27)

21 | P a g e

4.5 Study design

In a survey which has been distributed via thesistools.com 207 respondents were asked to choose under which delivery condition they would purchase a T-shirt. Eight choice sets were offered to each respondent which contained three different stimuli (delivery conditions). The choice sets are created with Sawtooth software SSI Web- 6.6.6. This study makes use of a fixed design, which means that respondents see the same questionnaire. However when one version of the questionnaire is being offered, too many choice sets have to be evaluated by the respondent which can lead to inaccurate answering. To offer less choice sets and make sure that respondents keep their attention to answering the questions two versions have been designed which leads to less choice sets to be evaluated per version. The respondents have to evaluate eight profiles, which consists of seven profiles used for estimation of the model and one hold out task to check the validity of the utilities estimated from the seven profiles, see appendix 1 questions 13-20 for the choice sets question 17 is the hold out task. The hold out task is thus “held out” of the utility estimation. The holdout task provides a sanity check to ensure that the model is working properly (Sawtooth SSI Web v6.6.6.). The respondents were randomly assigned to either version one of the questionnaire or version two. Furthermore some general question were asked in order to describe the sample see the questionnaire in appendix 1, questions 1-12

(28)

22 | P a g e

table 2. Meaning that the minimum efficiency level is above the threshold of 0.95 which is sufficient in order to get reliable results.

Attribute Freq. Actual Ideal Efficiency

Product availability: > 10 SKU available 14 (level has been deleted ) Product availability: Between 0-10 SKU available 14 0.3801 0.3780 0.9890 Product availability: Not available, will be delivered

when available again according provided delivery time

14 0.3866 0.3780 0.9557

Delivery time: 1 day 14 (level has been deleted)

Delivery time: 3 days 14 0.3815 0.3780 0.9813

Delivery time: 5 days 14 0.3837 0.3780 0.9701

Delivery costs: €0,00 14 (level has been deleted )

Delivery costs: €1,00 14 0.3845 0.3780 0.9663

Delivery costs: €4,95 14 0.3837 0.3780 0.9701

Price: €29,95 14 (level has been deleted)

Price: €32,95 14 0.3860 0.3780 0.9588

Price: €34,95 14 0.3802 0.3780 0.9884

Recommendation: 3 stars out of 5 14 (level has been deleted) Recommendation: 4 stars out of 5 14 0.3860 0.3780 0.9588 Recommendation: 5 stars out of 5 14 0.3802 0.3780 0.9884

Table 2: Design efficiency rates

4.6 Factor analysis.

(29)

23 | P a g e

4.7 OLS regression

(30)

24 | P a g e

5. Results

5.1 descriptive statistics

To have an idea what kind of respondents there are in the dataset some descriptive statistics are given. From the 207 respondents who participated 82 (39,6%) are men and 125 (60,4%) are women. The age of the respondents varies between 17 and 87 years old with an average of 32 year and 30% of the respondents consists of the age between 22-25 years. From the 207 respondents 170 (82,1%) live in a city and 37 (17,9%) live in a rural area. Table 3 gives an overview of the income distribution in the sample, an income below €33.000 occurs most in the sample. Most of the respondents have an HBO (43,5%) or WO (37,7%) education, which means a high education., table 4 gives the education levels in the sample.

Table 3: Income distribution

Table 5 gives an overview of the experience in online shopping. What can be observed from the data is that people have less experience in online apparel shopping than in overall experience of online shopping. This could imply that people who have purchased online are not so familiar with apparel shopping. Furthermore 45 (21,7%) respondents indicate that they never purchased a product online and that they are also not planning to do that in the nearby future and 162 (78,3%) respondents have purchased a product online or are planning to do so.

Experienced in online shopping Experience clothing Experience overall Frequency Percent Frequency Percent

Strongly disagree 50 24,2 13 6,5

Disagree 40 19,3 31 15,0

Somewhat disagree 22 10,6 24 11,6

Neither agree or disagree 28 13,5 29 14,0

Somewhat agree 33 15,9 52 25,1

Agree 27 13,0 42 20,3

Strongly agree 7 3,4 112 7,7

Total 207 100,0 207 100,0

Table 5: Experience in online shopping based on clothing and overall experience

Income Frequency Percent <€33.000 114 55,1

€33.000 26 12,6

>€33.000 56 27,1

unknown 11 5,3

Total 207 100,0

Education Frequency Percent

(31)

25 | P a g e

Table 6 gives an overview of the number of times the respondents buy clothes either online or offline, 44% of the respondents buys less than 6 times clothes per year, some people even do not buy clothes in a year. Furthermore the money spent on clothes is given in table 7, almost 78% spent less than € 100,- on clothing per month.

Table 6: Number of clothes purchased per year

Table 7: Money spent on clothes per month

Almost 50% of the respondents indicated that they enjoy shopping offline respective to 26% of the respondents who enjoy to shop online, table 8 gives an overview of the enjoyment of shopping online or offline.

Table 9 gives an overview of the rates that the respondents gave on the importance of the attributes. We asked the respondents to indicate to what extent the different attributes are important to them. What is observed is that price is the most important rated (75,9%) followed by delivery costs (57%) and delivery time (42,5%), somewhat lower rated attributes are product availability (36,7%) and recommendation (17,9%).

Money spent on clothes Frequency Percent

€ 0-50 63 30,4 € 51-75 55 26,6 € 76-100 43 20,8 € 101-125 28 13,5 € 126-150 8 3,9 € 151-200 7 3,4 >€ 200 3 1,4 Total 207 100,0

Clothes Frequency Percent

0 7 3,4 1-5 84 40,6 6-10 45 21,7 11-20 39 18,8 >20 32 15,5 Total 207 100,0

Enjoy Shopping Online Offline

Frequency Percent Frequency Percent

Strongly disagree 84 5,8 56 3,9

Disagree 203 14,0 77 5,3

Somewhat disagree 252 17,4 119 8,2

Neither agree or disagree 217 15,0 189 13,0

Somewhat agree 322 22,2 308 21,3

Agree 266 18,4 455 31,4

Strongly agree 105 7,2 245 16,9

Total 1449 100,0 1449 100,0

(32)

26 | P a g e

Product avail. Delivery time Delivery costs Recommendation Price

Freq. Percent Freq. Percent Freq. Percent Freq. Percent Freq. Percent

Strongly disagree 12 5,8 3 1,4 6 2,9 10 4,8 4 1,9 Disagree 33 15,9 10 4,8 4 1,9 22 10,6 2 1 Somewhat disagree 19 9,2 17 8,2 5 2,4 32 15,5 1 0,5 Neither agree/disagree 22 10,6 32 15,5 23 11,1 51 24,6 14 6,8 Somewhat agree 45 21,7 57 27,5 51 24,6 55 26,6 29 14 Agree 48 23,2 57 27,5 77 37,2 32 15,5 73 35,3 Strongly agree 28 13,5 31 15 41 19,8 5 2,4 84 40,6 Total 207 100 207 100 207 100 207 100 207 100

5.2 Choice Based conjoint analysis 5.2.1 Aggregate model

For estimating the parameters first the preference function of every attribute has to be determined. There are different sorts of preferences functions which are part-worth, quadratic and linear functions. Part-worth functions have to be estimated for each different level per attribute and a linear function uses one specific parameter for a specific attribute multiplied by the attribute’s value to arrive at part-worth value for each level. By default, variables containing values are treated as numeric and character variables are treated as nominal. In this conjoint analysis five attributes are included as noticed, product availability, delivery time, delivery costs, recommendation and price. At first a one class model has been used to estimate the parameters for each given attribute and its levels. In order to find out if parameters can be estimated as linear, the parameters are plotted to see if there exist a linear relationship between the levels of the investigated attributes, except for product availability, because this attribute is by default nominal and part-worth parameters have to be estimated. The parameters of the attributes that can be estimated as linear are recommendation and price, because the parameters show a linear relation as can be seen in figure 2. The parameters of delivery costs and delivery time have to be estimated as a part-worth function. When estimating the parameters linearly the degrees of freedom increase, because less parameters have to be estimated. The Bayesian Information Criteria (BIC), which is used to determine which models fits best, weights the fit and the parsimony of a model (the lower the BIC the better the model), it decreases when labeling attribute ‘recommendation’ (2152,31) as numeric followed by the attribute ‘price’ (2147,36) whereas from starting point all attributes estimated nominal had a BIC value of 2157,62. Table 10 gives an overview of the estimated parameters of the final aggregate model with only ‘recommendation’ and ‘price’ estimated as a linear relation.

(33)

27 | P a g e

The model performance of the estimated aggregate model can be calculated by the internal rate, meaning how many of the estimated choices match the observed choices. The internal hit rate of the aggregate model given in table 10 is 80,67%, which is a great improvement compared to predicting choices randomly, because if doing so the internal hit rate is expected to be 33,33 % (three options possible). The relative importance of the attributes for the aggregate model identifies how important a specific attribute is compared to other attributes.

Attributes Parameters Wald p-value

Product Availability

> 10 SKU available 0,4460 236,239 0,000

Between 0-10 SKU available 0,4667

Not available -0,9127 Delivery Time 1 day 0,5127 116,304 0,000 3 days -0,0941 5 days -0,4186 Delivery Costs € 0,00 0,7726 325,436 0.000 € 1,00 0,5764 € 4,95 -1,3489 Price -0,9524 274,511 0,000 Recommendation 0,2415 34,972 0,000

(34)

28 | P a g e

The ranges of the attributes are used for determining the relative importance. A range can be described as the difference between the highest and lowest parameters of the attribute and is divided, to calculate the relative importance, by the total sum of the ranges of attributes. For the aggregate model delivery costs (31%) are rated as most important attribute followed by price (28%) and product availability (20%), less important attributes are delivery time (14%) and recommendation (7%).

5.2.2. Interpreting the hypotheses on aggregate model

Product availability has a positive effect on the utility of the respondents, however there is almost no difference between levels ‘> 10 SKU available’ (0,4660) and ‘between 0-10 SKU available’ (0,4667), therefore people are indifferent on the level of SKU available. To be sure that there is no significant difference between these two levels the model has also been estimated with dummies instead of effects. Attribute level ‘>10 SKU available’ was coded as a dummy, subsequently the z-value for attribute level ‘between 0-10 SKU available’ has been interpreted to determine if there is a significant difference between the two levels. The z-value of attribute level ‘between 0-10 SKU available’ < 1,96 meaning that there is no significant difference between the two levels. Though there is a positive effect between whether a product is available or not and therefore H1a is accepted. The difference in utility between an out-of-stock situation and a situation with limited stock is larger than the difference between limited stock and full stock, meaning that H1b is accepted as well.

There is a clear negative significant effect for the delivery time and delivery costs, therefore H2 and H3a are accepted. Furthermore H3b is rejected because the difference between ‘free delivery’ and ‘€1,00’ is 0,1962 in utility and between ‘€1,00’ and ‘€4,95’ the difference is 0,4874 utility per euro delivery costs ((0,5764 + 1,3489)/3,95). So there is not a stronger effect between ‘free delivery’ and ‘€1,00’. H4 and H5 are accepted, there is a positive effect on the level of recommendation and there is a negative effect on utility for price.

5.2.3. Factor analysis for moderators (covariates)

(35)

29 | P a g e

case the value is 0,545, meaning that a factor analysis can be applied. Table 11 gives an overview of the components, as can be seen the ‘ratio1’ and ‘ratio2’ questions asked clearly belong to each other, the same holds for impatience and opinion. Now SPSS can save factor scores which are a linear combination of standardized variables. Now that we know the underlying dimensions there has to be tested for internal consistency of the factors. To test for internal consistency we use Cronbach’s alpha, if Cronbach’s alpha is higher than 0,6 the consistency is high enough to proceed with the factors. All Cronbach’s alphas are higher than 0,6 meaning that the factors are internal consistent, ‘rationality’ (0,732), ‘impatience’ (0,763) and ‘attaching value to opinion of others’ (0,706). In this way the moderators are created and a value is attached to each moderator per respondent. In the choice based conjoint analysis we can implement these moderators as covariates in order to test for the given hypotheses based on these moderators.

Component Impatience Opinion Ratio Cronbach’s alpha

ratio1 -0,073 0,075 0,881 0,732 ratio2 -0,071 0,033 0,861 impatience1 0,772 -0,037 -0,08 0,763 impatience2 0,892 0,067 -0,06 impatience3 0,796 0,077 -0,017 opinion1 0,102 0,85 0,085 0,706 opinion2 0,066 0,772 0,301 opinion3 -0,051 0,739 -0,168

5.2.4. Latent class analysis

Now that the preference functions are known and the moderators are calculated the number of latent classes can be determined. In order to determine the optimal cluster solution a choice based conjoint analysis is run with a range from one till eight segments. At first no covariates are included yet, because first there has to be decided on how many latent classes there are in the sample, best solution then is a three class solution with a BIC score of 2088,5204 in contrast to second best four-class solution with a BIC score of 2091,9864. In both three and four class solution the Wald statistic is significant with a p-value < 0,01, meaning that for each attribute the estimated parameters significantly differ from zero. Based on these facts a three or four class solution is preferred. However, active covariates will generally affect the definition of the classes and the part-worth and other parameters. Regarding the fact that the BIC scores do not differ that much in estimating the model for three or four class solution, the model is re-estimated with covariates for both classes, because active covariates will have an

(36)

30 | P a g e

effect on parameters. When estimating the model with covariates for three and four class solution some covariates are significant as can be seen in table 12. The next step is to re-estimate the model without the insignificant covariates. When pruning the covariates the only two significant covariates remaining are ‘Impatience’ and ‘Rational’ which can be seen in table 13. The BIC scores decrease for both classes, with the lowest for the four class solution.

Table 12: Covariates of 3 and 4 class solution

Table 13 Remaining significant covariates

If the parameters are interpreted of the three and four cluster solution then a four cluster solution is more preferred because of the Wald(=) statistic which tests if the effect of the attributes are different among classes. With three classes Wald(=) statistic for delivery time is 7,798 (p-value of 0,099) and for recommendation 4,445 (p-value of 0,11), meaning that there are no significant effect differences across segments with p-value < 0,01 see table 14. With a 4 class solution every Wald(=) statistic is significant, see table 14, with a p-value < 0,01, meaning that for every class there are significantly differences among effects of the five attributes. Furthermore the prediction error decreases in when choosing a four class (19,32%)

Covariates 3 class solution

BIC= 2215,2229

4 class solution BIC= 2279,9777

Wald p-value Wald p-value

Impatience 16,272 0,000 14,306 0,003 Rational 12,108 0,002 4,009 0,260 Age 2,378 0,300 1,320 0,720 Gender 15,351 0,000 7,785 0,051 Education 12,702 0,240 10,923 0,760 Residence 16,480 0,000 5,501 0,140 Income 9,260 0,160 16,825 0,051 Purchasedonline 17,516 0,000 12,438 0,006 NumbClothes 11,200 0,004 4,489 0,210 ShopOnlineEnjoy 17,852 0,000 11,679 0,009 ShopOfflineEnjoy 1,840 0,400 10,198 0,017 ExperienceClothing 1,997 0,370 1,105 0,780 ExperienceOverall 9,476 0,009 9,321 0,025 MoneySpentClothes 13,197 0,001 8,884 0,031 Opinion 4,730 0,094 13,954 0,003

Covariates 3 class solution BIC= 2085,1556

4 class solution BIC=2084,8318

Wald p-value Wald p-value

Rational 8,220 0,016 18,477 0,000

(37)

31 | P a g e

instead of a three class (22,57%) solution. Finally a four class solution is preferred, because of the lower BIC score with covariates included, significant differences between classes and lower prediction error (higher internal hit rate).

The parameters can now be estimated for a four class solution with active covariates ‘Rational’ and ‘Impatience’, attributes ‘recommendation’ and ‘price’ set as numeric and the remaining attributes ‘product availability’, ‘delivery time’ and ‘delivery costs’ set as nominal preference function. The other covariates are included as inactive covariates as well and can be used to describe the segments, they will have no effect on parameter estimations. Table 15 gives an overview of the estimated parameters.

5.2.5. Interpreting the segments and hypothesis

Attributes 3-class solution BIC= 2085,1556

4 class solution BIC=2084,8318

Wald p-value Wald(=) p-value Wald p-value Wald(=) p-value ProductAvailability 111,939 0,000 22,533 0,000 132,538 0,000 33,336 0,000 DeliveryTime 117,110 0,000 7,798 0,099 107,639 0,000 16,880 0,009 DeliveryCosts 98,503 0,000 52,675 0,000 129,914 0,000 48,971 0,000 Recommendation Price 36,166 116,136 0,000 0,000 4,445 32,277 0,110 0,000 64,104 111,487 0,000 0,000 59,068 27,282 0,000 0,000

Table 14 Wald statistic and p-values of three and four class solution

Attributes Class size Segment A 0,29 Segment B 0,29 Segment C 0,26 Segment D 0,16 ProductAvailability > 10 SKU available 1,0844* 0,2802 0,2587 0,8667* Between 0-10 SKU available 1,2916* 0,3331* 0,4794* 0,3016 Not available -2,3759* -0,6132* -0,7381* -1,1682* DeliveryTime 1 day 0,7776* 0,6202* 0,4074* 0,9727* 3 days 0,0678 -0,339 0,3144* -0,1016* 5 days -0,8454* -0,2812 -0,7218* -0,8711* DeliveryCosts € 0,00 2,4155* 1,5983* 0,4411* 0,1394 € 1,00 1,5093* 0,9594* 0,4955* 0,4831* € 4,95 -3,9248* -2,5577* -0,9366* -0,6226* Price Recommendation -1,7533* -2,3033* -0,7204* -0,8062* -0,2465 0,1255 0,9662* -0,2300 Covariates Rational Impatience -0,7950 0,3410 0,4799 -0,0259 -0,0284 -0,5667 -0,2433 0,8384 *z-value > 1.96 and therefore significant at a 5% level

(38)

32 | P a g e

All four segments prefer product availability above not available products. There is a positive effect for all segments when there are available products, therefore H1a can be

supported. The difference in utility between an out-of-stock situation and a situation with limited stock is larger than the difference between limited stock and full stock for all segments, meaning that H1b is accepted as well. For segment A, B and C the preferred option

is ‘between 0-10 SKU available’ and for segments D the preferred attribute level is the one with ‘>10 SKU available’. Product availability has the biggest influence on the utility for both positive and negative effects in segment A.

For all four segments there is a negative effect of delivery time, because the longer it takes till the product is delivered the lower the utility gets. Only for segment B the people are indifferent between a delivery time of three or five days, because of insignificant parameters, but there is still a negative effect, because a one day delivery time effects the utility with 0,6202 instead of a zero effect for three or five days. To conclude it can be said that H2 is supported.

For H3a to be true there has to be a negative effect on the level of utility for delivery costs.

Only Segments A and B support this, because most preferred in both segments is a ‘free delivery’ followed by ‘€1,00’ and ‘€4,95’ and so a negative effect occurs. Remarkable for segments C and D is that they prefer delivery cots of ‘€1,00’ above the one of ‘free delivery’ and so H3a cannot be supported for those segments The utility of segment A is strongest

affected by delivery costs and smallest affected for segment D. H3b stated that there is a

stronger negative effect between ‘free delivery’ costs and ‘€1,00’ than between ‘€1,00’ and ‘€4,95’, this hypothesis cannot be supported for all segments. Because for segment A the difference in effect is -0,8252 (2,4155-1,5093) for ‘free delivery’ and ‘€1,00’ and the difference in effect for ‘€1,00’ and ‘€4,95’ is -1,8114 ((1,5093 + 3,9246)/3). For segment B the difference in effect is 0,6389 (1,5983-0,9594) for ‘free delivery’ and ‘€1,00 and the difference in effect for ‘€1,00’ and ‘€4,95’ is -1,1724 ((0,9594 + 2,5577)/3) so there is a stronger negative effect in both segments between ‘€1,00’ and ‘€4,95’ than for ‘free delivery’ and ‘€1,00. Segments C and D prefer delivery costs of ‘€1,00’ and so there is not even a negative effect from ‘free delivery’ to ‘€1,00’.

(39)

33 | P a g e

for price (-0,7204) on utility. H4 is supported, because all four segments have a negative effect on the utility as price increases. Remarkable for the provided level of rating is that only for one segment the parameter is significantly different from zero, which is the one for segment C (0,9662). Segment C shows a positive effect for the recommendation by previous customers. The higher the level of rating the higher the utility gets, segments A and D show even a negative effect for recommendation, however since these parameters are not significantly different from zero these segments are indifferent for the provided recommendation, same holds for segment B. Only for segment C H5 is supported and in the other segments people are indifferent about recommendation and so H5 cannot be supported for those segments.

The relative importance of the attributes for the given segments are given in figure 3. It shows how important attributes are for segments A-D. Figure 3 shows that the segments value different aspects of the delivery process. For segment A ‘delivery costs’ are the most important aspect of the delivery process. Segment B attaches much value to the price of the given product, closely followed by ‘delivery costs’. Segment C attaches most value to the recommendation provided by others, but has no clear second most important attribute. This indicates that for those people in segment C the attributes are evenly important and no clear preference is stated. This can be a sign of rational thinking, meaning that respondents take all different attributes into account for best decision making. People in segment C are most rational compared to other segments (0,4799), see table 16 for covariates. The product availability is for Segment D the most important attribute followed by delivery time. This indicates that they prefer a fast delivery time and are not keen on an out-of-stock situation. People in segment D are most impatient (0,8384), see table 16 for covariates, compared to other segments A-C. The strongest significant negative effect for a delivery time of five days (-0,8711) is found in segment D and therefore H6b can be supported, impatient people have a

stronger effect on the level of utility for ‘delivery time’. H6a is not supported, because

impatient people do not have the lowest utility for an out of stock situation in Segment D. Segment A has strongest effect for an out of stock situation and are less impatient than segment D.

Moderators Segment A Segment B Segment C Segment D

Rational -0,7950 0,3410 0,4799 -0,0259

Impatience -0,0284 -0,5667 -0,2433 0,8384

Optimal delivery conditions for an online retailer

Master Thesis, Thomas Wiegerinck

University of Groningen

28-06-13

Optimal delivery conditions for an online retailer

MANAGEMENT SUMMARY

PREFACE

Table of Contents

1. INTRODUCTION

2. LITERATURE REVIEW

3. CONCEPTUAL MODEL

4. METHODOLOGY

5. Results