Online grocery shopping: the use of pickup points
What is the Dutch customer willing to pay for the use of a pickup point when
ordering groceries online?
J.M. (Jojanneke) Dierssen University of Groningen Faculty of Economics and Business MSc thesis: Marketing Intelligence
January 11, 2016
Landsberger Allee 92 10249 Berlin +31 614534998
j.m.dierssen@student.rug.nl Student number: 2590875
MANAGEMENT SUMMARY
Online grocery shopping is becoming increasingly popular. More and more supermarkets are offering this service to meet the modern consumers’ needs. With online grocery shopping, there is not only a delivery option: customers are also able to pick up their groceries at a so-called pickup point. The aim of this research is to get more insights in this specific area, since limited literature is available. The main goal is to find the price customers are willing to pay for picking up their online ordered groceries. This could provide managers insights to develop more targeted strategies, that could increase the sales volume of their online channel.
With the help of a literature review and choice-based conjoint analysis, four important factors affecting the willingness to pay for pickup points have emerged. First, assortment size is affecting the willingness to pay in such a way, that a large assortment is preferred over a small assortment. This has to do with convenience for the customer and being able to order everything that is needed in once. Second, smaller distances are preferred over longer distances. Third, processing time is not that important for customers. However, a short processing time is in general preferred. Finally, there are differences between preferences for in-store pickup points and stand-alone pickup points. The most important difference is that the willingness to pay is higher for stand-alone pickup points (€9.87 on average) than for in-store pickup points (€2.75 on average).
The content of this thesis is as follows: chapter 1 introduces online grocery shopping, and specifically pickup points, in an extended way. Current numbers of online grocery shopping are given, in order to create a clear view of this service. Chapter 2 provides a theoretical framework, consisting of definitions of key concepts, existing literature, hypotheses and the conceptual model. The methodology of this research is given in chapter 3, followed by the results in chapter 4. Finally, chapter 5 consists of theoretical implications, managerial implications, limitations and suggestions for future research.
PREFACE
This thesis is written under somewhat unusual circumstances. A month before the kick-off of the graduation process, I moved to Berlin, out of love for the city. For years I looked forward to the first and best opportunity to take this step. A semester without lectures, just a thesis, that should be it! However, living in a dream it was in the beginning kind of hard to focus completely on my thesis. Luckily I realized soon that this thesis is very important and I was very motivated to deliver a fine piece of work. So after months of hard work, I present you my thesis with pride.
Discussing with my peers Aneta Rabljenovic and Jan Nies was incredibly helpful, so I want to thank them for the good and successful collaboration. Furthermore, I really want to thank my first supervisor, Erjen van Nierop, for giving many and extensive feedback and giving me the opportunity to have meetings through Skype. Finally, I want to thank my family and friends for their support and endless faith in me.
TABLE OF CONTENTS
1 INTRODUCTION ... 7
2 THEORETICAL FRAMEWORK ... 9
2.1 Definitions ... 9
2.2 Pickup points ... 9
2.3 Drivers and hurdles ... 11
2.4 Influencing variables ... 12
Assortment ... 12
2.4.1 Distance to pickup point ... 13
2.4.2 Processing time ... 14 2.4.3 Price of products ... 14 2.4.4 2.5 Customer groups ... 15 2.6 Conceptual model ... 16 3 METHODOLOGY ... 17 3.1 Depth interviews ... 17 3.2 Conjoint analysis ... 18 Data collection ... 19 3.2.1 Attributes ... 19 3.2.2 Experimental design ... 20 3.2.3 Model specification ... 21 3.2.4 4 RESULTS ... 23 4.1 Depth interviews ... 23
Respondent who never bought online before ... 23
4.1.1 Respondent who bought online before, but not groceries ... 23
4.1.2 Respondent who bought groceries online before (low experience) ... 23
4.1.3 Respondent who bought groceries online before (high experience)... 23
4.1.4 Conclusion ... 24
4.1.5 4.2 Dataset ... 24
4.3 Descriptives: all respondents (before weighing) ... 24
Demographic characteristics ... 24
4.3.1 4.4 Descriptives: no-choicers excluded + weighing ... 25
Demographic characteristics ... 25
Pickup points ... 27
4.4.3 Conclusion ... 28
4.4.4 4.5 Conjoint analysis: latent class analysis for in-store pickup points ... 28
Model form ... 28
4.5.1 The initial model from 3.2.4 is adjusted into the following model: ... 29
Model selection ... 29
4.5.2 Model fit ... 30
4.5.3 Parameters and importance per segment ... 30
4.5.4 Covariate parameters per segment ... 31
4.5.5 Profiles ... 32
4.5.6 Segment 1: Convenience seekers ... 33
4.5.7 Segment 2: Bargain hunters ... 33
4.5.8 Segment 3: Short-distance shoppers ... 34
4.5.9 4.6 Conjoint analysis: latent class analysis for stand-alone pickup points ... 35
Model form ... 35 4.6.1 Model selection ... 36 4.6.2 Model fit ... 36 4.6.3 Parameters and importance per segment ... 37
4.6.4 Covariate parameters per segment ... 38
4.6.5 Profiles ... 38
4.6.6 Segment 1: Experienced convenience seekers ... 39
4.6.7 Segment 2: Curious bargain seekers ... 40
4.6.8 4.7 Hypotheses testing ... 40
5 CONCLUSIONS & RECOMMENDATIONS ... 41
5.1 Theoretical implications ... 41
5.2 Managerial implications ... 43
5.3 Limitations and future research ... 44
REFERENCES ... 45
1 INTRODUCTION
A new era has dawned: information, technology, rapid innovation and new market needs keep on coming. A lot of these developments have to do with the Internet. The numbers of Internet usage in the Netherlands are impressive. 96% of all Dutch people uses the Internet, whereas 71% of the total Dutch population shops online (Syndy, 2015). The food sector does not elude these developments. Supermarkets respond fast and smart to it, causing the newest service called ‘e-grocery’. This phenomenon refers to the online ordering of groceries by customers.
An important motivation for supermarkets to add an online alternative is that the service can increase customer satisfaction and loyalty. Also, it can help to retain existing customers (IGD, 2012; Zhang, Farris, Irvin, Kushwaha, Steenburgh and Weitz, 2010), which might lead to increased chain sales and profits (Mulpuru et al., 2012). At this point, 6,4% of the Dutch population buys groceries online regularly with an average expenditure of €1500 per year (Actiepagina, 2015). Overall, only 1,5% of the food products in the Netherlands is purchased over the Internet (NRCQ, 2015). Reasons for this small percentage are, among others, a high supermarket density (26 supermarkets per 100.000 inhabitants) and the mindset of consumers, since they want to feel the products (Marketingfacts, 2015). However, this online food retail market is expanding due to an increasing number of online players, better coverage and growing order possibilities for Dutch consumers (Gorczynski and Kooijman, 2015). The same authors found that 80% of the Dutch households prefers to have groceries delivered to their home, since ‘convenience’ is a key driver behind online grocery shopping. As for pickup points, the main reason to favor them over home delivery is the flexibility for the online customer to collect the ordered groceries at their preferred times. A very important finding in the research of Gorczynski and Kooijman is that 25% of the households is willing to pay to pick up groceries (on average €2.88), while 60% is willing to pay for home delivery (€3.87).
As mentioned above, consumers have two options when ordering their groceries online: they let their groceries deliver at home or they pick up the groceries at a so-called ‘pickup point’. Over the last two years, the pickup point network in the Netherlands has seen extreme growth of almost 680%, increasing from 45 pickup points in January 2013 to more than 350 in December 2014.
research which is consumers’ willingness to pay (WTP) for picking up their groceries, resulting in the following research question:
What amount of money are consumers willing to pay for picking up their groceries and which variables are of influence?
In order to get a clear answer, this question is divided into four sub-questions. By answering these questions the overall conclusion will follow.
Ramus and Nielsen (2005) and Morganosky and Cude (2000) mentioned in their study several reasons for customers to shop their groceries online. What are these reasons today, and what are hurdles for online grocery shopping?
1. What are drivers and hurdles for the use of pickup points from the customer’s perspective? There are many variables that could have an effect on the willingness to pay. These effects can be different for the two types of pickup points, in-store pickup points and stand-alone pickup points. The first type of pickup point can be found in supermarkets, the latter can be found at several places, for example at highways or fuel pumps.
2. Which variables have an effect on the WTP for pickup points?
3. What are differences in the effects for in-store pickup points and stand-alone pickup points?
According to a study into willingness to buy groceries online, although in a slightly different setting, it will be examined whether a distinction can be made between the effects on the WTP for different customer groups (Hansen, 2008). These customer groups are:
- Customers that never bought online before
- Customers that bought something online before, but not groceries - Customers that bought groceries online before
4. What kind of distinction can be made between the effects on the WTP for different customer groups?
There is not only a practical relevance of this paper, but also a theoretical relevance. There is little research about e-grocery, in particular with regard to pickup points. This is due to the fact that the service is relatively new and is still developing itself. More research in the field is needed, in order for managers to know how they can serve their customers best and make their companies grow in an innovative way at the same time. The aim of this is study is to contribute to this research area.
2 THEORETICAL FRAMEWORK
This section starts with the definitions of the most important key terms in this research. After that, hypotheses will be formulated based on existing literature. As mentioned before, this is a relatively new field of research and therefore, this chapter also contains some practical studies. Finally, the hypotheses are combined into the conceptual model at the end of this section, which creates the basis for the remainder of this research.
2.1 Definitions
There are some key concepts that will return throughout this research. The most important definitions are given below:
- E-grocery stands for electronic grocery shopping. It is a way of buying food and household products using a web-based shopping service (Hallsworth, 2006), or as defined by Verhoef and Langerak (2001), ordering grocery from home in an electronic way and having them delivered at one’s house. However, when ordering groceries online, one can choose for home delivery as well as for the use of pickup points.
- A pickup point is a prearranged place where people can go to collect their groceries. As mentioned before, there are two types of pickup points (Syndy, 2015). In-store pickup points can be found in supermarkets, stand-alone pickup points can be found at several places, for example at highways or fuel pumps.
- Willingness to pay (WTP) is defined as the maximum amount of money a customer is willing to pay for a product or service (Koschate-Fischer, Stefan and Hoyer, 2012), or to avoid something undesirable. The price of a good transaction will be any point between a buyer’s willingness to pay and a seller’s willingness to accept.
2.2 Pickup points
marketing costs and costs with regard to complaints are lower for pickup points than for home delivery. However, the respondents within the research are willing to pay much more for home delivery (€7) in contrast to pickup points (€4). A reason for this might be that they link the fee to their baskets sizes, which turns out to be higher on average for home delivery than for pickup points.
FIGURE 1
The economics of home delivery versus pickup points (Galante et al., 2013)
The aforementioned two kinds of pickup points have each their own advantages and disadvantages. The first category of pickup points, in-store pickup points, does not require a lot of investments and staff is already present. Also, there is no duplicate stock-keeping and a low floor space requirement. However, there is a double workload for replenishment, in-store congestion and there are faster out of stocks. The second category, stand-alone pickup points, is more flexible in terms of business hours and locations. It is also more convenient for consumers in terms of reachability and increases the number of possible touch-points with consumers. However, there are high initial investments, which cause the need for more stakeholders such as investors (Syndy, 2015). All advantages and disadvantages of each type are presented in table 1.
when it comes to pickup points, since it seems that ‘time-saving’ as driver for online grocery shopping is optimally utilized when the pickup point itself is on the way to the customers’ destination. On the other hand, with regard to in-store pickup points, these are more convenient when the consumer has forgotten to order something. They can easily go in the store and buy the forgotten products. However, since this is an ‘in case of’-scenario, it can be expected that stand-alone pickup points are more attractive in the end.
H1: The willingness to pay is higher for stand-alone pickup points compared to in-store pickup points.
TABLE 1
Two types of pickup points: advantages and disadvantages (Syndy, 2015)
Advantages Disadvantages
In-store pickup point Small initial investments
No duplicate stock-keeping
Low floor space requirement
Flexible staff planning
Double workload for replenishment
Faster out of stocks
In-store congestion
Lower consumer willingness to use compared to home delivery
Stand-alone pickup point Flexible: longer business hour and more locations
Convenient for consumers
Increases the number of touch-points with consumers
High initial investments
Lower consumer willingness to use compared to home delivery
Involvement of more stakeholders
2.3 Drivers and hurdles
There are different drivers as well as hurdles in the minds of customers when it comes to e-grocery. Several studies investigated these drivers and hurdles, all showing that often the same factors emerge.
Ramus and Nielsen (2005) mention a few advantages of e-grocery as compared to conventional grocery shopping. They appoint product range, price and convenience as biggest advantages of online grocery shopping. Convenience as important driver is also reflected in the study of Kinsey and Senauer (1996). They make clear that retailers have the potential to create value along two dimensions of convenience. One of these dimensions is reducing the amount of time required to complete the shopping task. According to Kinsey and Senauer (1996), the ultimate time-saving convenience may be home shopping. The study of Corbett (2001) is in line with this finding. He found that convenience as well as time saving factors are the primary drivers for buying groceries online. Also Szymanski and Hise (2000) support convenience as biggest advantage. They found that perceptions of convenience are the most important factor in e-satisfaction assessments.
are primary reasons, but also physical or constraint reasons that made it difficult to shop at physical grocery stores. Morganosky and Cude asked their respondents to compare the time spent shopping online with the time spent shopping in the store. Many of the respondents commented that the real time savings were a result of not traveling to and from the store, rather than a decrease in shopping time. The respondents argue that the process of ordering groceries online takes time as well.
Actiepagina (2015) researched the reasons for online grocery shopping as well, with again convenience and saving time as primary drivers. The respondents of this study mention that it is easier to avoid impulse purchases when shopping groceries online. Also, almost all respondents save money when shopping groceries online by using online discount codes. Thereby, it is easier to search for special offers, since there is a better overview and it is easy to view the offers in a few clicks. However, several hurdles are mentioned as well. The respondents want to pick their own products at expiry date or freshness. Also a lot of people find it more enjoyable and sociable to go to the store. In addition, some people consider only briefly in advance what they want to eat and thus the products they need. This makes it more convenient for those people to go to the store instead of ordering online.
Finally, an article by ‘Tijdschrift voor Marketing’ (Marketingonline, 2013) reveals several drivers and hurdles for people to make use of a pickup point. Again, saving time is mentioned as the most important advantage of a pickup point. Also deciding about the point of time to pick up the groceries is an advantage, as compared to staying at home for the delivery. For the majority of the respondents in the article is not picking your own fresh products a reason to not make use of the online service. Also missing offers and new products are considered to be annoying. Table 2 summarizes the outcomes of the listed studies.
TABLE 2
Summary drivers and hurdles of online grocery shopping
Drivers Hurdles
Convenience People want to pick own (fresh) products Saving time Loss of recreational/social aspect Product range Needs known on short notice
Price Missing offers/new products
Physical/constraint reasons
2.4 Influencing variables
There are many variables that can possibly affect the willingness to pay for picking up groceries at a pickup point. This paragraph will discuss the most important variables.
Assortment 2.4.1
need to respect very short delivery times can be reasons to restrict online assortments for some product categories. Oppewal and Koelemeijer (2005) and Sloot, Fok and Verhoef (2006) suggest that larger assortments tend to be preferred over smaller ones, because they offer more choice flexibility and enhance feelings of autonomy. Besides, Briesch, Chintagunta and Fox (2009) found that assortments are more important than retail prices in store choice decisions. They also found that the number of brands offered in retail assortments have a positive effect on store choice. This suggests that a supermarket needs a very large assortment with an endless amount of different products. However, this is not the case at all. Wathieu, Cameron, Chattopadhyay, Wertenbroch, Drolet, Gourville, Muthikrishnan, Novemsky, Ratner and Wu (2002) mentioned in their article about consumer control and empowerment, restricting a consumer’s choice set to fewer alternatives makes it easier to make a decision and leads to greater satisfaction with the decision. Another finding is that the main problem consumers have with an extensive array of options in the pre-decision stage, is the confusion that too many alternatives can cause. These findings make clear that an assortment should be extensive, but that there is a certain limit.
Melis, Campo, Breugelmans and Lamey (2015) state that assortment attractiveness is an important factor of store loyalty and purchase intention, also in an online shopping context. Therefore, the online environment facilitates store comparison on assortment attractiveness. In these terms there is a big advantage for online supermarkets, since limited shelf space is no obstacle. This allows supermarkets to offer a wider assortment with products outside the standard range (Foodlog, 2014). Altogether, it can be concluded that assortment size is a crucial factor and can have a positive effect on the WTP, but that the effect can become negative when the assortment is too big.
H2: Assortment size has a positive effect on the WTP for pickup points, but has decreasing returns.
Distance to pickup point 2.4.2
Travel time and transportation costs can discourage households from visiting offline stores and encourage them to visit the online store (Bell, Ho and Tang, 1998; Chintagunta, Chu and Cebollada, 2012; Forman, Ghose and Goldfarb, 2009), whereby they let their bought stuff deliver at home. However, there are several reasons for consumers to make use of a pickup point. For example, the consumer drives along the pickup point when going home from work, and does not want to pay extra money for home delivery. In this case, travel time and transportation costs barely play a role.
used to drive relatively longer distances. This is very positive for retailers, since home delivery is extra costly in rural areas. Furthermore, Pate and Loomis (1995) found in their study about the effect of distance on willingness to pay values that willingness to pay declines as distance increases: the greater the distance to the pickup point, the smaller the willingness to pay.
Overall, the majority does not want to drive more than 5 kilometers to a pickup point, according to a research of Marketingonline (2013). Deloitte (2014) found that when people have to drive specially to a pickup point, so not when they are already on their way, they are willing to drive a maximum of 11 minutes to pick up their groceries. For these reasons, it is expected that the distance to the pickup point has a negative effect on the willingness to pay for pickup points.
H3: The distance to the pickup points has a negative effect on the WTP for pickup points. Processing time
2.4.3
Supermarkets in general have multiple dimensions to compete on. Price is of course very important, but consumers pay also attention to other dimensions of the product or service, for example quality, various elements of customer service and delivery speed in an online environment. It is well known that speed and consistency of delivery time are the most important elements of customer service (Sterling and Lambert, 1989; Wu, Webster and Yang, 2012) and that the company should excel on these factors. Otherwise, it is very easy for a consumer to switch to the competitor.
As mentioned before, saving time is one of the biggest drivers for online grocery shopping. With time as an important factor, the duration from ordering to the point of time to pick up the groceries is important as well. With regard to home delivery, most retailers deliver the groceries within 24 hours (RTL Nieuws, 2015). For pickup points the duration is shorter: supermarket employees only need to pack the groceries. It can be assumed that people who order their groceries online want to pick it up as soon as possible, or at the moment they want. Therefore it is important that the groceries should be packed as soon as possible.
H4: Processing time has a negative impact on the WTP for pickup points. Price of products
2.4.4
According to Chu, Arce-Urriza, Cebollada-Calbo and Chintagunta (2008), heavy online shoppers are more price sensitive in the online channel than in the offline channel, whereas light online shoppers are less price sensitive in the online channel than in the offline channel. This difference is larger for food products than non-food products. However, Degeratu, Rangaswamy and Wu (2000) and Andrews and Currim (2004) found that online consumers are less price sensitive when shopping groceries online than offline.
outweigh the disadvantages for customers (Campo and Breugelmans, 2015) and they rather go to an offline store. Also, just as assortment, price is a determining factor when it comes to offline store choice (Melis et al., 2015).
In the Netherlands, the prices for products in the supermarkets and the online shop are in general the same. Also online and offline offers are generally similar. For this reason it was decided not to include price of the products as variable in the remainder of the study. If the prices online and offline are the same, price as such will not have an effect on the willingness to pay when ordering groceries online.
2.5 Customer groups
As mentioned before, this research will include ‘customer groups’ as a moderator. The aim of this is to find out whether there are different effects on the willingness to pay between several groups. The division is adopted from a research of Hansen (2008), where he investigated relations of consumer personal values, attitude, social norm, perceived behavioral control and willingness to buy groceries online. Hansen distinguished the following three groups of consumers:
Group 1: Customers that never bought online before
Group 2: Customers that bought something online before, but not groceries Group 3: Customers that bought groceries online before
Melis et al. (2015) used a moderator ‘experience’ in their research about the moderating effect of online experience in online store choice. They used two levels to measure experience, based on the number of orders: low experience and high experience. The results of this study show that when online grocery shopping experience increases, multi-channel shoppers’ focus shifts from a comparison within a chain across channels to a comparison across chains within the online channel. Research by Actiepagina (2015) shows that a third of the Dutch people cannot imagine ever doing online grocery shopping. This group mainly consists of people over 50, who are relatively less familiar with the internet and online shopping than young people. Another finding shows that families with young children use e-grocery most often. Also Marketingonline (2013) found out that families with children make most use of pickup points. They also found that elderly people do not like the idea of online grocery shopping and pickup points, since they are less familiar with online shopping and still prefer to visit a store.
Based on the aforementioned studies, it can be assumed that there are differences between the effects on the willingness to pay for the different customer groups.
2.6 Conceptual model
The following hypotheses were formulated based on existing literature:
H1: The willingness to pay is higher for stand-alone pickup points compared to in-store pickup points.
H2: Assortment size has a positive effect on the WTP for pickup points, but has decreasing returns.
H3: The distance to the pickup points has a negative effect on the WTP for pickup points. H4: Processing time has a negative effect on the WTP for pickup points.
H5: There are differences between the effects on the willingness to pay for different customer groups.
Based on the theory and hypotheses, the conceptual model can be drawn (figure 2). FIGURE 2
Conceptual model H2: Assortment size
H3: Distance to pickup point
3 METHODOLOGY
This chapter explains the data collection method for this research. The two methods discussed are depth interviews, specifically laddering, and conjoint analysis.
3.1 Depth interviews
The first sub-question was ‘What are drivers and hurdles for the use of pickup points from the customer’s perspective?’. Paragraph 2.3 describes different theories, which answers the question. Therefore, it is decided not to include the question in quantitative research. Instead, it has been decided to conduct a qualitative research, which provides insights and understanding of the problem setting (Malhotra, 2010), in order to gain more in-depth answers for this question. Several depth interviews are held to see if the found answers are in line with the drivers and hurdles of the Dutch population at this point of time.
A depth interview can be defined as an ‘’unstructured personal interview which uses extensive probing to get a single respondent to talk freely and to express detailed beliefs and feelings on a topic’’ (Stokes and Bergin, 2006). There are several techniques which can be applied during a depth interview. To answer this particular sub-question, it is decided to use the technique ‘laddering’. Laddering provides a way to probe into consumers’ deep underlying psychological and emotional reasons that affect their purchasing decisions (Malhotra, 2010). In other words, it is a way of measuring consumers’ means-end chains that makes use of semi-structured personal interviews (Gutman, 1982). The means-end value chain is presented in figure 3. The idea behind this chain is that the respondents proceed from product attributes (A) to user values (V) by means of functional and psychosocial consequences (C).
FIGURE 3
Means-end value chain (Gutman, 1982)
The technique that is used is ‘evoking a situational context’. This technique works best for this particular case, because the respondents are providing associations while thinking of a realistic occasion in which they would use the service (Reynolds and Gutman, 1988).
(people that bought groceries online before) is split up in two groups: low experience and high experience, to see if there are any differences in answers. Thus, there will be a total of four depth interviews:
1. Respondent who never bought online before
2. Respondent who bought online before, but not groceries
3. Respondent who bought groceries online before (low experience) 4. Respondent who bought groceries online before (high experience) 3.2 Conjoint analysis
The other three sub-questions were as follows:
2. Which variables have an effect on the WTP for pickup points?
3. What are differences in the effects for in-store pickup points and stand-alone pickup points?
4. What kind of distinction can be made between the effects on the WTP for different customer groups?
A conjoint analysis is used as main method to get answers on these questions. With a conjoint analysis, it is possible to measure consumer preferences and to reveal underlying motives for their actions (Eggers and Sattler, 2011). The respondents who are participating are asked to evaluate different ‘service bundles’ in terms of their desirability. These service bundles are combinations of attribute levels, which are the characteristics of the services (Malhotra, 2010).
The conjoint analysis is used to determine the WTP for the use of pickup points. One way to do this is to detect the price at which an alternative changes from being the most preferred to equally preferred to a second best option. This price then is the maximum WTP for that alternative, because at higher prices, another option would be chosen instead (Eggers and Sattler, 2011). A conjoint analysis matches this research well, since the goal is to discover what people are willing to pay for the use of pickup points and what variables (attributes) have an impact on this.
There are several methods to use within a conjoint analysis: traditional conjoint, adaptive conjoint and choice-based conjoint (Orme, 2003). In this research, the choice-based conjoint analysis is used. This is a more real and representative method than the other methods (Hair et al., 2014). In this case, it gives repeatedly three different service bundles and an option of not choosing any of the presented bundles by including a ‘no-choice’ option.
Data collection 3.2.1
The data is acquired via both the Internet and face-to-face surveying, in order to reach all the different kinds of respondents. Face-to-face surveying takes place in front of both stand-alone and in-store pickup points. The program used for the online survey is Qualtrics Research Suite, which is an online survey software tool. The survey was reviewed by a professional in pricing and choice models and tested by several respondents before sending the survey in order to ensure data validity and reliability. According to these tests, the survey was modified to make it as easy and straightforward as possible for the respondent to fill in.
Attributes 3.2.2
The survey starts with two demographic questions (age and gender), followed by a question about which supermarket is visited most by the respondent. The next questions are about online shopping, online grocery shopping, how often the respondent shops his/her groceries online and how often he/she picks up the online ordered groceries at a pick up point. These last questions are asked to define to which customer group the respondent belongs. The attributes (i.e. independent variables) are selected based on the research in chapter two of this study. All attributes contain three levels in order to prevent the number-of-levels effect, which occurs when the number of levels are not distributed equally across attributes. This would result in higher importance for attributes with more levels.
Assortment size. Because the attribute levels should have a concrete and unambiguous meaning, this attribute is presented by means of pictures combined with verbal cues. The pictures are used to illustrate different sizes of assortments, since this can be hard to imagine when only reading a number. As example, the category ‘tea’ will be used. Three different levels are determined: small (ten different kinds of tea), medium (20 kinds of tea) and large (30 kinds of tea):
Distance to pickup point. In paragraph 2.4.2 it became clear that most people do not want to drive more than 5 kilometers to pick up their groceries. Besides, people are not willing to drive longer than 11 minutes to a pickup point. It has been decided to determine 3 levels based on kilometers instead of minutes, since this is easier to imagine when filling in the survey. The next three levels are determined: 2.5, 5, 7.5 kilometers.
Processing time. Processing time is the time it takes from ordering the groceries online till the moment the groceries are ready to be picked up. The following levels are determined: within 2 hours, same day, next day. Although these levels are not equal what might occur the number-of-levels effect, they are chosen since realistic number-of-levels are more important.
not want to pay at all for the service. Therefore, the following three levels of pickup costs has been chosen: €0, €2 and €4.
No-choice. The no-choice option is one of the major benefits of the choice based conjoint analysis, because it increases realism. It gives respondents the opportunity to reject all three alternatives, when the prices of those alternatives are too high (Eggers and Sattler, 2011). In this study, the no-choice option is picked when a customer is not using the service of picking up groceries, but is going to the supermarket and collect the groceries himself.
Table 3 summarizes the attributes with their corresponding attribute levels. TABLE 3
Attribute levels conjoint analysis
Attributes Attribute levels
Assortment size Small Medium Large
Distance to pickup point 2.5 km 5 km 7.5 km
Processing time <2 hours Same day Next day
Costs €0 €2 €4
No choice option
Experimental design 3.2.3
The respondents are presented with two different experimental designs: choice sets for stand-alone pickup points and choice sets for in-store pickup points. Prior to each part, it has been made clear what pickup points are and a distinction between the two types is made. After that, it is highlighted about what kind of pickup point the subsequent questions are. In order to prevent that it takes too long for the respondents, the fractional factorial design is used in both cases. This fractional factorial design is a subset of the full factorial design, which shows all possible attribute level combinations. To ensure efficiency, each level is displayed an equal number of times (balanced) and each level combination appears an equal number of times (orthogonal). Each design consists of six choice sets, each choice set consists of 3 attribute bundles and a no-choice option. Both designs are randomized, meaning that each respondent will see a different selection of choice sets. Thus, when setting up the survey, 24 different choice sets are entered. The respondent sees however only 12 choice sets. Figure 4 shows one of the choice sets, in this case specifically for an in-store pickup point. Appendix I shows the complete survey (in Dutch), including the explanations but with 2 choice sets, used for this study.
Model specification 3.2.4
With conjoint analysis, choices are based on overall utilities of alternatives. The utility of respondent n for alternative i looks as follows:
𝑈𝑛𝑖 = 𝑉𝑛𝑖 + 𝜀𝑛𝑖
with
V = systematic utility component, rational utility Ԑ = stochastic utility component, error term.
An assumption within the conjoint analysis is that goods and services are combinations of attributes. The respondents attach part-worth utilities to each attribute. The systematic utility of respondent n for alternative i is the sum of part-worth utilities:
𝑉𝑛𝑖 = ∑ 𝐾
𝑘=1 𝛽𝑛𝑘𝑋𝑖𝑘
with
k = number of attributes
𝑋𝑖𝑘 = dummy indicating the specific attribute level of product i
𝛽𝑛𝑘 = utility of consumer n for attribute k
Furthermore, a multinomial logit (MNL) model is used, since the dependent variable can exhibit multiple states. By implementing the attributes into the previous model, the initial model looks as follows:
𝑉𝑛𝑖= 𝛽1𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒𝑆𝑚𝑎𝑙𝑙 + 𝛽2𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒𝑀𝑒𝑑𝑖𝑢𝑚+ 𝛽3𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒𝐿𝑎𝑟𝑔𝑒 + 𝛽4𝑖𝐷𝑖𝑠2.5𝑘𝑚 + 𝛽5𝑖𝐷𝑖𝑠5𝑘𝑚
+ 𝛽6𝑖𝐷𝑖𝑠7.5𝑘𝑚+ 𝛽7𝑖𝑇𝑖𝑚𝑒<2ℎ𝑜𝑢𝑟𝑠 + 𝛽8𝑖𝑇𝑖𝑚𝑒𝑆𝑎𝑚𝑒𝐷𝑎𝑦 + 𝛽9𝑖𝑇𝑖𝑚𝑒𝑁𝑒𝑥𝑡𝐷𝑎𝑦
+ 𝛽10𝑖𝐶𝑜𝑠𝑡𝑠€0+ 𝛽11𝑖𝐶𝑜𝑠𝑡𝑠€2+ 𝛽12𝑖𝐶𝑜𝑠𝑡𝑠€4
where
𝑉𝑛𝑖 = utility for the chosen option
𝛽1𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒𝑆𝑚𝑎𝑙𝑙 = dummy variable for small assortment for alternative i
𝛽2𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒𝑀𝑒𝑑𝑖𝑢𝑚 = dummy variable for medium assortment for alternative i
𝛽3𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒𝐿𝑎𝑟𝑔𝑒 = dummy variable for large assortment for alternative i
𝛽4𝑖𝐷𝑖𝑠2.5𝑘𝑚 = dummy variable for 2.5km distance for alternative i
𝛽5𝑖𝐷𝑖𝑠5𝑘𝑚 = dummy variable for 5km distance for alternative i
𝛽6𝑖𝐷𝑖𝑠7.5𝑘𝑚 = dummy variable for 7.5km distance for alternative i
𝛽7𝑖𝑇𝑖𝑚𝑒<2ℎ𝑜𝑢𝑟𝑠 = dummy variable for <2 hours time for alternative i
𝛽8𝑖𝑇𝑖𝑚𝑒𝑆𝑎𝑚𝑒𝐷𝑎𝑦 = dummy variable for same day time for alternative i
𝛽10𝑖𝐶𝑜𝑠𝑡𝑠€0 = dummy variable for €0 costs for alternative i
𝛽11𝑖𝐶𝑜𝑠𝑡𝑠€2 = dummy variable for €2 costs for alternative i
𝛽12𝑖𝐶𝑜𝑠𝑡𝑠€4 = dummy variable for €4 costs for alternative i
Finally, based on this utility function, the probability that alternative i is chosen out of a given choice set of J is determined by:
4 RESULTS
This chapter reports the results of the depth interviews, descriptive analyses as well as the conjoint analysis. After each section there is a brief conclusion. In addition, the chapter ends with a general conclusion.
4.1 Depth interviews
Four depth interviews have been held with different respondents, each representing one of the four previous mentioned groups.
Respondent who never bought online before 4.1.1
This person is an elderly retired woman. She has a lot of children and grandchildren who visit her often. She always makes sure she has enough to eat and drink at home. The woman does her own grocery shopping, but sometimes she is helped by one of her grandchildren on Wednesday afternoon. She cannot imagine to buy her groceries online, since she likes it to walk through the supermarket and talk to the other villagers. Also, she wants to determine the freshness of the products herself and to take the longest life products.
Respondent who bought online before, but not groceries 4.1.2
This is a student who is daily studying at the library or university. He visits the supermarket every day. Often he shops groceries with his roommates, also they cook and eat together multiple times a week. He thinks there is no advantage for him in ordering groceries, especially not when he has to pay for using the service. He sees it as extra proceedings and longer waiting times, while the nearest supermarket is around the corner of his house. Also, he only knows what he wants to eat when he is already in the supermarket and wants to see the products first.
Respondent who bought groceries online before (low experience) 4.1.3
This person is a woman with two small children. She works part-time. She does a few times a week her groceries and tries to get as much as possible groceries on Saturday for the whole week. She ordered her groceries a few times online before, but she does not want to drive around for it. However, when the pickup point is on the route, the distance to her house does not matter. She ordered groceries only occasional when she was giving a party, because for example crates of beer and soft drinks would be quite heavy for her.
Respondent who bought groceries online before (high experience) 4.1.4
himself. Pickup points would thus not make sense for him. The groceries are sometimes ordered by his wife as well and mainly occupies food and drinks for the employees of the company.
Conclusion 4.1.5
The depth interviews were conducted with four totally different persons in order to get various answers. The results are in line with the results of the literature review. Mostly the same hurdles as well as drivers are mentioned by the interviewees. The attributes costs, time, distance and assortment are underpinned by the results. Furthermore, since these four persons are different in characteristics and preferences, it is tried to let all age categories and genders be represented within the survey. This way, there will be as much as possible insights obtained into the preferences concerning the use of pickup points when ordering groceries online.
4.2 Dataset
A total of 440 respondents participated in the survey, of which 311 respondents completed the survey. Only the completed surveys are analyzed. The dataset is adapted in several ways. Since the conjoint analysis questions are entered as pictures in the survey, the attributes and choices had to be transformed into variables: four variables for the attributes, one variable for the no-choice option and a dummy variable which indicates the choice of the respondent. This way it is easy to see which respondents have only chosen the no-choice option, who are deleted from the dataset. After removing those respondents, there are 152 respondents left. Furthermore, the dataset is weighed based on the variable ‘age of the respondent’, since this variable showed imbalances, to create a more representative age distribution. Also, a new variable ‘groups’ has been created, whereby the variables ‘ever bought something online before’ and ‘ever bought groceries online before’ are merged to one variable. With this new variable, it is easy to detect the effects of the different customer groups. With these adaptations and included variables the dataset is prepared to be analyzed with Latent Gold. Finally, two different models estimated: one model for in-store pickup points and one model for stand-alone pickup points.
4.3 Descriptives: all respondents (before weighing)
This paragraph shows concise descriptives of the dataset, consisting of all completed surveys and including the no-choicers, before weighing.
Demographic characteristics 4.3.1
TABLE 4
Demographic characteristics of all respondents
Characteristic Category Percentage
Gender Men 27% Women 73% Age <18 3% 18-29 55% 30-39 18% 40-49 10% 50-59 8% 60-69 4% >70 2%
When looking at the data, a lot of these respondents have repeatedly chosen only the 'no choice' option within the conjoint analysis part of the survey. These respondents are not that interesting for this research since they show no interest in pickup points at all, so it has been decided to exclude them from the conjoint analysis. The group of only no-choice respondents consists of 31% men and 69% women and are mostly 18-29 years old (57%). The remaining group of respondents will be analyzed more in depth first, before moving to the conjoint analysis.
4.4 Descriptives: no-choicers excluded + weighing
Weighing is a frequently used method to correct the representativeness. With this procedure, a weight is given to each respondent. People in groups that are underrepresented in the sample, get a weight greater than 1. People in overrepresented groups get a weight below one.
Demographic characteristics 4.4.1
TABLE 5
Demographic characteristics respondents
CBS 2014 ConsumentenTrends 2014 Before weighing After weighing
Gender Men 49,5% 29% 32% 35% Women 50,5% 71% 68% 65% Age <18 20% - 4% 13% 18-34 21% 24% 67% 40% 35-54 28% 41% 24% 38% >55 31% 35% 5% 9%
Figure 5 displays the supermarket preferences of the respondents. It is obvious that Albert Heijn and Jumbo are the most preferred supermarkets, which largely corresponds to numbers of the union Consumentenbond (2014). According to this union, Albert Heijn was the most popular supermarket in the Netherlands in 2014, followed by Jumbo and Lidl at a shared second place.
FIGURE 5
Supermarket preferences
With regard to the customer groups, there are very large differences within the group of respondents (table 6). The first group, people who never bought something online, is heavily underrepresented with a percentage of 5, which is much less than the 29% mentioned by Syndy (2015). The third group, respondents who bought groceries online before, is remarkable large. This is due to the fact that all no-choicers are deleted from the dataset, since this group consisted of 27% of the respondents before leaving out all no-choicers. Leaving out these respondents results in a bigger group with people who have at least an open mind towards online groceries
TABLE 6
Customer groups distribution
Customer group Percentage
1: Never bought something online 5%
2: Bought something online before 44%
3: Bought groceries online before 51% 0 10 20 30 40 50 Albert Heijn
Jumbo Lidl Aldi Coop Plus Other
Online grocery shopping 4.4.2
As mentioned before, 51% of the respondents has ever ordered groceries online before. Figure 6 shows the proportions of respondents’ total grocery trips replaced with an online order in the past half year. As can be seen, most people (14,5%) tried it a few times, but did not do it on a regular basis. This group consists of somewhat more women (54,5%) than man (45,5%) and shops in particular at Albert Heijn (77,3%).
FIGURE 6
Percentages online grocery ordering in the past half year of the total supermarket visits
Pickup points 4.4.3
The use of pickup points is strongly divided within this group of respondents. Figure 7 shows that many people who order their groceries online do not use pickup points at all (30,7%) or use it
(almost) all the time (35,5%). The respondents who do not use pickup points are in particular women (68,4%) and usually shop at Albert Heijn (52,6%). The respondents who do use pick up points
(almost) all the time are also in particular women (63,6%) and usually shop at Albert Heijn (68,2%). Thus, there is not a big difference between moderate and frequently users of a pickup point service.
FIGURE 7
Percentages use of pickup points versus home delivery
0 5 10 15 20 Perc en ta ge s
Proprortion of total grocery trips replaced with online order
Conclusion 4.4.4
The dataset was adjusted by means of weighing in order to get a more representative dataset. The weighing procedure was conducted with the age variable. The weighed dataset is used in Latent Gold, so the descriptives were mainly about this set. The most important results showed that there are few people who shop their groceries online on a regular basis. Also, when people order their groceries online, people either let the groceries deliver at home most often or they pick the groceries up themselves most often.
4.5 Conjoint analysis: latent class analysis for in-store pickup points
This first part of the conjoint analysis takes the in-store pickup points into account. Model form
4.5.1
Attribute formats are not necessarily the same for all attributes. The separate part-worth (nominal) parameters of the attributes are plotted on an aggregate level to visually check whether a linear relationship is appropriate (figure 8). This model is estimated without the covariates.
FIGURE 8 Part-worth parameters
As can be seen, all attributes might have a linear relationship. To test for this, the model is re-estimated with each time different relationships and again only the covariates in it. Table 7 shows these results, starting from all attributes part-worth (nominal) to all attributes linear. Model four has the lowest BIC (1166.4323) as well as CAIC (1172.4323), which means that a model with assortment, distance and time as linear, and thus costs as nominal, suits best.
-0,8 -0,6 -0,4 -0,2 0 0,2 0,4 0,6 0,8 1 Sm all Me d iu m La rge 2,5 5 7,5 < 2 h o u rs Sam e d ay N ext d ay 0 2 4
Util
ity
TABLE 7
Re-estimation of model with different relationships
LL BIC(LL) CAIC(LL) #par p-value Hit rate
Model 1 All attributes nominal -568.6000 1176.6942 1185.6942 9 <.001 49%
Model 2 Assortment linear -568.8842 1172.8745 1180.8745 8 <.001 49%
Model 3 Assortment + distance linear
-569.8496 1170.4170 1177.4170 7 <.001 49%
Model 4 Assortment + distance + time linear
-570.0514 1166.4323 1172.4323 6 <.001 47%
Model 5 All attributes linear -575.4832 1172.9077 1177.9077 5 <.001 47,7%
The initial model from 3.2.4 is adjusted into the following model:
𝑉𝑛𝑖= 𝛽1𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒 + 𝛽2𝑖𝐷𝑖𝑠 + 𝛽3𝑖𝑇𝑖𝑚𝑒 + 𝛽4𝑖𝐶𝑜𝑠𝑡𝑠€0+ 𝛽5𝑖𝐶𝑜𝑠𝑡𝑠€2+ 𝛽6𝑖𝐶𝑜𝑠𝑡𝑠€4
where
𝑉𝑛𝑖 = utility for the chosen option
𝛽1𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒 = linear variable for assortment for alternative i
𝛽2𝑖𝐷𝑖𝑠 = linear variable for 2.5km distance for alternative i
𝛽3𝑖𝑇𝑖𝑚𝑒 = linear variable for <2 hours time for alternative i
𝛽4𝑖𝐶𝑜𝑠𝑡𝑠€0 = dummy variable for €0 costs for alternative i
𝛽5𝑖𝐶𝑜𝑠𝑡𝑠€2 = dummy variable for €2 costs for alternative i
𝛽6𝑖𝐶𝑜𝑠𝑡𝑠€4 = dummy variable for €4 costs for alternative i Model selection
4.5.2
A latent class analysis is performed since it is expected that consumers differ in their preferences. To figure out what is the best number of classes, five different models are performed. The best number of classes can be determined on the basis of different criteria. The most important information criteria in terms of penalties are the BIC and the CAIC. They have higher penalties for complexity and are appropriate for larger sample sizes, because the penalty grows with the sample size. The model in table 8 is estimated with all covariates in it. As can be seen, the BIC and CAIC are continually increasing. Furthermore, when looking at the different models in Latent Gold, it gives an estimation warning from the 5-Class Choice, since the degrees of freedom become negative here. This is due to a small number of respondents in proportion with the number of parameters and the use of a replication weight. Therefore, models with five or more classes should not be considered.
TABLE 8 Model selection
BIC(LL) CAIC(LL) #par p-value Hit rate Class.Err.
2-Class Choice 1077.1207 1104.1207 27 <.001 59% 0.0179
3-Class Choice 1106.6433 1154.6433 48 <.001 66% 0.0346
4-Class Choice 1150.4867 1219.4867 69 <.001 69% 0.0234
To find out which number of classes gives the best solution, the class sizes within each solution are compared. The first model has two classes of respectively 62% and 38%; the second model has three classes of respectively 37%, 33% and 30%; the third model has four classes with respectively 37%, 29%, 23% and 11%. Although the 3-Class Choice model has the highest classification error, it has the best distribution. Furthermore, the 4-Class Choice has a rather small class of 11%, so it has been decided to pick model three to continue with.
Model fit 4.5.3
To see how this 3-Class Choice model performs, first a comparison with the aggregate model has been made. Table 9 shows that the 3-Class Choice model performs better on all criteria (BIC=1106.6433; CAIC=1154.6433, R2adj=0.62763). Also, this 3-Class Choice model has a high hit rate
of 66% compared to the aggregate model’s hit rate of 47%, meaning that 66% of the observations is predicted correctly by the model.
TABLE 9
Comparison with aggregate model
BIC(LL) CAIC(LL) #par p-value Hit rate R2adj Class.Error
3-Class Choice 1106.6433 1154.6433 48 <.001 66% 0.6273 0.0346
Aggregate model 1166.4323 1172.4323 6 <.001 47% 0.5672 0.0000
Besides a comparison with the aggregate model, the goodness of fit of the model is defined the likelihood ratio test. The Likelihood Ratio test is used to see if the estimation model predicts better than the NULL model. The chi-squared distribution has 33 degrees of freedom, so the critical value is (α=5%) = 47.400. The Chisq test statistic is as follows:
LL(0) = 80*12*ln(1/4) = -1,330.842 Chisq = −2*(-1,330.842-(-448.0035)) = 1765.677
p(1765.677) < 0.0001.
This means that there is a difference between the estimation model and the NULL model, and thus that the estimation model has a good overall fit.
Parameters and importance per segment 4.5.4
accounts more or less the same: a negative effect in class 1 and 2 and a quite small positive effect in class 3. With regard to costs, the parameters for price become more negative in each class if the price increases. So, all three classes prefer lower costs over higher costs.
TABLE 10
Parameters per segment
Attributes Class 1 (37%) Class 2 (33%) Class 3 (30%) Wald p-value Wald(=) p-value Std.Dev.
Assortment 0.6391 0.2208 0.5939 22.4698 <0.001 3.3167 0.190 0.1878 Distance -0.3977 0.0460 -0.4092 59.0965 <0.001 36.9376 <0.001 0.2109 Time -0.7790 -0.1563 0.1717 13.4804 <0.001 10.7672 0.005 0.3981 Costs 0 2 4 1.3720 0.1750 -1.5469 1.1002 -0.7283 -0.3719 0.5373 -0.0634 -0.4739 107.882 <0.001 15.1325 0.004 0.3440 0.3861 0.5444 No choice -1.2114 -3.8521 -0.4739 39.3193 <0.001 5.6064 0.061 1.1005
Table 11 shows the relative importance of each attribute within the three different segments. With this relative importance, it is possible to compare between the three segments.
TABLE 11
Relative importance of the attributes per segment
Segment 1 Ranking 1 Segment 2 Ranking 2 Segment 3 Ranking 3
Assortment 17% 4 16% 2 36% 2
Distance 26% 2 8% 4 39% 1
Time 20% 3 11% 3 6% 4
Costs 37% 1 65% 1 19% 3
The previous paragraph made clear that the attribute assortment has, according to the Wald(=) statistic, more or less the same influence on all three segments. Furthermore, distance is most important in segment 3 (39%), but is also quite important in segment 1 (26%). With regard to time, this attribute is the least important attribute of all, but most important in segment 1 (20%). Finally, the attribute costs is the most important attribute of all attributes, especially in segment 2 (65%).
Covariate parameters per segment 4.5.5
TABLE 12
Covariate parameters per segment
Covariates Segment 1 Segment 2 Segment 3 Wald P-value
Gender Man Woman -1.2528 1.2528 1.5909 -1.5909 -.3381 .3381 6.7760 0.034 Age <18 18-29 30-39 40-49 50-59 60-69 0.8417 -.8702 -1.8550 -.2663 .1938 1.9560 -.8919 -1.9406 .5071 -3.6509 -1.9444 7.9207 .0502 2.8108 1.3479 3.9171 1.7506 -9.8767 11.2118 0.34
Supermarket Albert Heijn Jumbo Lidl Aldi Coop Plus Other -.0817 -.8717 .0763 .0178 -.9289 .9771 .8111 -1.9378 -1.0570 -2.2931 -3.1426 3.3421 7.6544 -2.5660 2.0195 1.9287 2.2168 3.1249 -2.4132 -8.6315 1.7549 5.8103 0.92
Group Never ordered something Ordered something Ordered groceries -.5660 1.0889 -.5229 -5.1823 1.6340 3.5483 5.7483 -2.7229 -3.0254 6.8288 0.15 Profiles 4.5.6
The segments will be described based on the profiles, including all attributes and the covariate gender, as well as the willingness to pay. Table 13 shows each segments’ profile. Although the differences for the attribute assortment are not significant, will it be used to describe the segments.
TABLE 13
Profiles based on attributes and covariate gender
Profile 1 (37%) Profile 2 (33%) Profile 3 (30%)
Assortment Small Medium Large 16% 29% 55% 26% 33% 41% 16% 30% 54% Distance 2.5 5 7.5 66% 25% 9% 30% 33% 37% 67% 24% 9% Time <2 hours Same day Next day 60% 27% 13% 39% 33% 28% 28% 33% 39% Costs €0 €2 €4 74% 22% 4% 72% 12% 16% 52% 29% 19% Gender Man Woman 6% 94% 67% 23% 35% 65%
TABLE 14
Relative willingness to pay per attribute
Segment 1 Segment 2 Segment 3
Assortment 0.4379 -0.3000 1.1748
Distance -0.2725 0.0625 -0.8093
Time -0.5337 -0.2124 0.3396
No choice -0.8300 -5.2332 -5.0086
TABLE 15
Absolute willingness to pay per segment
Segment 1 Segment 2 Segment 3 In-store pickup point
WTP €0.99 €4.27 €3.25 €2.75
Segment 1: Convenience seekers 4.5.7
This segment is the largest segment with 37% of all respondents in it. The segment consists almost entirely of women (94%) and is characterized by being very clear in its preferences. First of all, this segment prefers a large assortment (55%) and a short distance of 2.5 km (66%). Also, the people in this group do not want to wait too long, preferably less than two hours (60%) or want to pick up their groceries the same day (27%). Finally, most people in this segment do not want to pay for the use of a pickup point at all (74%). All these preferences together make it obvious that this segment consists of convenience seekers: convenience against no price is wanted. The parameters in table 10 support this: distance (β=-0.3977) and time (β=-0.7790) have negative parameters, assortment a positive parameter (β=0.6391), whereas the parameters of costs decrease when the costs increase. Finally, the willingness to pay of this segment is €0.99, which is remarkably lower than the WTP’s for the other segments. Table 16 gives an overview of this segment.
TABLE 16
Segment 1: Convenience seekers
Segment 1: Convenience seekers
Mostly women (94%)
Large assortment preferred (55%) Ranking 4
Short distance of 2.5 km (66%) Ranking 2
Short processing time of less than two hours (60%) Ranking 3
Costs preferred to be none (74%) Ranking 1
WTP €0.99
Segment 2: Bargain hunters 4.5.8
that the people in this segment find low costs the most important aspect of this service. The parameters are supporting this: assortment (β=0.2208), distance (β=0.0460) and time (β=-0.1563) have rather small parameters, whereas the parameters for costs are decreasing when going from €0 to €2 (β=1.1002, β=-0.7283). Finally, the willingness to pay of this segment is €4.27, which is the highest of all segments. Table 17 gives an overview of this segment.
Table 17
Segment 2: Bargain hunters
Segment 2: Bargain hunters
Mostly men (67%)
Assortment plays no large role Ranking 2
Distance plays no large role Ranking 4
Time plays no large role Ranking 3
Costs preferred to be none (72%) Ranking 1
WTP €4.27
Segment 3: Short-distance shoppers 4.5.9
This segment is the smallest segment with 30% of the respondents in it. The group mainly consists of women (65%). First, a large assortment is quite important in this segment (54%). With regard to the attribute distance, there is a clear preference for the shortest distance (67%). Distance is also the most important attribute within this segment (39%). Furthermore, the preferences for time are quite spread (28%, 33%, 39%) and is the least important attribute within this segment (6%). For costs goes that the lower is better (52%, 29%, 19%), but these preferences are somewhat less manifest than within the other segments. Altogether it can be concluded that the respondents in this segment go for the convenience of a short distance. The parameters are as follows: assortment has a positive parameter of β=0.5939, distance has a negative parameter of β=-0.4092, time a small positive parameter of β=0.1717 and the parameters for costs are decreasing (β=0.5373, 0.0634, β=-0.4739). Finally, the willingness to pay of this segment is €3.25. Table 18 gives an overview of this segment.
Table 18
Segment 3: Short-distance shoppers
Segment 3: Short-distance shoppers
Mostly women (65%)
Large assortment preferred Ranking 2
Distance is very important Ranking 1
Time plays no large role Ranking 4
Costs preferred to be none (52%) Ranking 3
4.6 Conjoint analysis: latent class analysis for stand-alone pickup points
This paragraph has the same structure as the previous paragraph, but is now focused on the stand-alone pickup points.
Model form 4.6.1
Also here, the separate part-worth parameters are plotted for the attributes to visually check whether a linear relationship is appropriate (figure 9). Again, this model is estimated without the covariates.
FIGURE 9 Part-worth parameters
All attributes might have a linear relationship. The model is re-estimated with different relationships to check this. Although the hit rate is lowest in model 5 (45,7%), this model with all attributes linear performs best (BIC=1203.3363; CAIC=1208.3363), since the BIC and CAIC keep decreasing as more and more attributes are included as linear (table 19).
TABLE 19
Re-estimation of model with different relationships
LL BIC(LL) CAIC(LL) #par p-value Hit rate
Model 1 All attributes nominal -590.0583 1219.6109 1228.6109 9 <0.001 45,9%
Model 2 Assortment linear -589.9732 1215.0524 1223.0524 8 <0.001 45,9%
Model 3 Assortment + distance linear
-590.1921 1211.1020 1218.1020 7 <0.001 45,9%
Model 4 Assortment + distance + time linear
-590.2598 1206.8491 1212.8491 6 <0.001 45,9%
Model 5 All attributes linear -590.6975 1203.3363 1208.3363 5 <0.001 45,7% -1 -0,8 -0,6 -0,4 -0,2 0 0,2 0,4 0,6 0,8 1 Sm all Me d iu m La rge 2,5 5 7,5 < 2 h o u rs Sam e d ay N ext d ay 0 2 4
Util
ity
The initial model from 3.2.4 is adjusted into the following model:
𝑉𝑛𝑖= 𝛽1𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒 + 𝛽2𝑖𝐷𝑖𝑠 + 𝛽3𝑖𝑇𝑖𝑚𝑒 + 𝛽4𝑖𝐶𝑜𝑠𝑡𝑠
where
𝑉𝑛𝑖 = utility for the chosen option
𝛽1𝑖𝐴𝑠𝑠𝑆𝑖𝑧𝑒 = linear variable for assortment for alternative i
𝛽2𝑖𝐷𝑖𝑠 = linear variable for 2.5km distance for alternative i
𝛽3𝑖𝑇𝑖𝑚𝑒 = linear variable for <2 hours time for alternative i
𝛽4𝑖𝐶𝑜𝑠𝑡𝑠 = linear variable for €0 costs for alternative i Model selection
4.6.2
Also here five different models including all covariates are estimated. In this model (table x), the BIC and CAIC are continually increasing. Again, an estimation warning is given from a 5-Class solution due to the small number of respondents and the use of the replication weight. Therefore, models with five or more classes should not be considered. The 2-Class solution gives groups of 61% and 39%; the 3-Class solution gives groups of 50%, 30% and 20%; the 4-Class solution gives groups of 39%, 25%, 19% and 17%. Based on this distribution of percentages across the different classes, a 2-Class choice has been chosen. The BIC, CAIC and Classification Error are here the lowest as well. Furthermore, a 2-Class solution is convenient in terms of manageability. An overview of the information criteria is given in table 20.
TABLE 20 Model selection
BIC(LL) CAIC(LL) #par p-value Hit rate Class.Err.
2-Class Choice 1109.3971 1134.3971 25 <.001 57,8% 0.0165 3-Class Choice 1114.9531 1159.9531 45 <.001 64,8% 0.0323 4-Class Choice 1158.8010 1223.8010 65 <.001 64,1% 0.0433 5-Class Choice 1204.7642 1289.7642 85 66,7% 0.0289 Model fit 4.6.3
To check the goodness of fit, a comparison with the aggregate model has been made. Table 21 shows that the 2-Class choice scores better on all criteria (BIC=1109.3971; CAIC=1134.3971, R2adj=0.6056).
This model has a hit rate of 57,8% compared to the aggregate model’s hit rate of 45,7%, meaning that 57,8% of the observations is predicted correctly by the model.
TABLE 21
Comparison with aggregate model
BIC(LL) CAIC(LL) #par p-value Hit rate R2adj Class.Error 2-Class Choice 1109.3971 1134.3971 25 <0.001 57,8% 0.6056 0.0165
