What drives Loyalty Program Participation? Choice-based Conjoint Analysis in the Restaurant Industry

(1)

What drives Loyalty Program Participation?

Choice-based Conjoint Analysis in the Restaurant

Industry

Msc Marketing Intelligence/Management

Master Thesis

University of Groningen

Faculty of Economics & Business

Januari 14, 2018

Supervisor

P.C. Verhoef

Author

Steven Visser (S2802430)

Frederik Hendrikstraat 33-I

(2)

(3)

3

ABSTRACT

Loyalty programs have made a surge in the current consumer environment in order to stimulate customer loyalty. Many studies have focussed on the elements that constitute loyalty program success in terms of fostering loyalty and increasing firm performance. However, this study was performed to transfer loyalty programs into an industry where loyalty programs are currently very scarce: The restaurant industry. Subsequently, the potential benefits of a loyalty program in the restaurant industry can only be obtained if a firm knows how to execute the first step in the design: How to create a loyalty program which effectively drives program participation? This report implements a choice-based conjoint analysis in the restaurant industry (n=147). The respondents were sampled within a 3-week period to test which combination of loyalty program attributes provides the highest utility and therefore, maximizes the probability that one joins the program. Thereafter, a latent class analysis was performed to account for difference among customers of a restaurant.

The results indicate that personal data investment, participation exclusivity, participation efforts, program benefits and program duration are all significant drivers of potential program participation. Program benefits is by far the most important driver, which accounts for 50% of the perceived program utility. More specifically, a program that offers utilitarian benefits such as discounts on food and drinks are highly favoured over hedonic and symbolic rewards. Personality characteristics such as age, purchase frequency and customer spending were found to be insignificant in determining customer loyalty participation intention.

In general, joining a restaurant loyalty program is favoured over not joining a restaurant loyalty program if it was available. This study therefore (1) proves the potential of implementing a restaurant loyalty program that is successful in attracting participants and (2) guides managers in the process of designing an optimal loyalty program applicable in the restaurant industry.

(4)

(5)

5

1. INTRODUCTION

Customer loyalty has been an important topic that has gained wide interest from both academic and managerial practitioners. (Morgan & Hunt, 1994; Grönroos, 1995; Meyer-Waarden, 2008). The conclusions of the studies have resulted in a widely recognized importance of loyal customers. Therefore, many firms spend vast amounts of resources on creating customer loyalty with the goal of achieving higher levels of customer retention, customer satisfaction and share of wallet (Bolton, Kannan & Bramlett, 2000). A second justification why numerous firms want to increase or expand customer loyalty is that customer loyalty has a positive influence on long-term financial performance (Anderson, Fornell & Lehmann, 1994; Reichheld & Sasser, 1989; Martinez & del Bosque, 2013). Furthermore, loyal customers are more likely to engage in positive word-of-mouth behaviours and spend extra money in a specific service operation (Ladhari, Brun & Morales, 2008; Tepeci, 1999, Yang & Peterson, 2004) In simple terms, a high-level relationship quality fostering customer loyalty can provide companies with a competitive edge that is difficult to imitate (Roberts, Varki & Brodie, 2003).

The power of loyal customers is also becoming increasingly important in the restaurant industry. Recently, many restaurants shifted focus from a short-term casual customer approach towards a long-term customer relationship and loyalty approach (Hyun, 2010). An example is restaurants placing personal comment cards on the dining table. (Kotler, Bowen, & Makens 1998). Afterwards, this data is inserted into a database which is used to foster loyalty by providing incentives such as rewards frequent dinners or personalized offers. (Kasavana & Knutson 2000). Furthermore, restaurant managers are showing increased understanding that loyal customers are very important to increase competitive power (Meng & Elliot, 2008), and are adapting marketing strategies to achieve increased loyalty (Mattila, 2001; Jang & Matilla, 2005; Hyun, 2010). A lot of larger chain-restaurants are spending extra staff and marketing resources into relationship quality managers (Hyun, 2010).

(8)

8

in more than one program (Berry, 2013; Berman, 2006; Kim, Lee, Wu, Choi & Johnson, 2013). An example of a loyalty program is the AH Bonuskaart, which rewards the customer with personalized and non-personalized discounts on groceries. A second example in the hotel industry is Marriot’s Rewards program. This loyalty program lets guests save points at each stay or purchase within the hotel, which can be redeemed for bonuses such as a free stay or dinner.

Generally, loyalty programs function as a tool to enlarge customer loyalty by providing increased satisfaction and more value to certain customers (Leenheer, van Heerde, Bijmolt & Smidts, 2007; Bolton, Kannan & Bramlett, 2000; Demoulin & Zidda, 2008). These goals are sought-after by providing customers with monetary advantages (e.g. discounts or coupons) and non-monetary rewards such as exclusive loyalty program member events or a priority table reservation system (Demoulin & Zidda, 2009).

Next to the benefits that loyalty programs give to customers, applications of loyalty programs are also useful from a firm’s perspective. Companies can use loyalty programs as a tool to (1) enhance customer retention (Leenheer et al., 2007), (2) increase existing loyalty (Meyer-Waarden, 2007) and (3) collect and analyse data about purchase behaviour (Kumar & Shah, 2004). Recent studies have shown that loyalty programs, when implemented correctly, are effective in their purpose and enable retailers to achieve and increase customer loyalty (Meyer-Waarden, 2007; Noordhoff, Pauwels & Odekerken-Schröder, 2004). At the same time, companies that offer loyalty reward programs believe that their programs have a positive effect on customer evaluations and behaviour (Bolton, Kannan, & Bramlett, 2000).

(9)

9

program are primarily U.S.-based restaurant chains focussing on casual dining, and usually exclude fine-dining restaurants (Matilla, 2005; Tepeci, 1999; Hyun, 2010).

This could imply that by not implementing a loyalty program, many restaurants are currently missing out on opportunities to enhance and increase highly-valued customer loyalty. Shapiro & Vivian (2000) support the potential of loyalty programs for the restaurant industry by stating that investing in customer loyalty programs is especially important when customers face low switching costs, which is typically the case in the restaurant industry (Chen & Hitt, 2002; Jones, Mothersbough & Beatty, 2002)

To be able to implement and access the full potential of a loyalty program in the restaurant industry, managers must know how an effective loyalty program should be designed. And in order to accomplish the aforementioned, it is important to understand the most salient preferences and motivations of customers to participate in such a program (Peltier, Schibrowski & Davis, 1998; Sharp, 1998; De Wulf, Odekerken-Schröder, De Canniére & van Oppen, 2003). Previous customer loyalty studies have investigated an almost inexhaustible list of aspects regarding customer loyalty programs. For example, the effect of loyalty programs on consumer behaviour (Nunez & Drèze, 2006; Bolton & Kannan, 2000), the application and purpose of data gathered within loyalty systems (Leenheer et al., 2002), the effects of customer loyalty on customer satisfaction and its subsequent benefits (Gustafsson, Johnson & Rose, 2005; Hennig-Thurau & Klee, 1997; Rust & Zahornik, 1993) and the crucial components of loyalty programs in order to foster loyalty (Downing & Uncles, 1997; Oliver, 1999). However, understanding of the attributes that stimulate of loyalty program participation is currently very underdeveloped. While Leenheer, van Heerde & Bijmolt (2007), Meijer-Waarden (2008) and De Wulf et al., (2003) have initiated research on this topic in a retail setting, the outcomes are often not congruent based on methodological differences (Leenheer et al., 2007). Furthermore, no evidence of similar research has been found in the restaurant industry specifically.

1.1 Research Objectives

(10)

10

2000; De Wulf et al., 2003). In short, the author wants to prevent managers from designing a loyalty programme that theoretically fosters loyalty, but nobody wants to join. Therefore, this research will analyse the effect of different loyalty program attributes on the likelihood of participating in a loyalty program. The study builds upon the equity-based model proposed by De Wulf et al., (2003) and applies it in the restaurant industry. Thus, the main research question is:

Which loyalty program attributes are important for customers to participate in a restaurant loyalty program?

By answering the research question, the author can derive the following findings:

- What is the optimal design for a loyalty program for restaurants that is attractive for customers?

- What do customers want to invest in, and obtain from, restaurant loyalty programme participation?

- Does restaurant loyalty program participation depend on personal characteristics?

1.2 Academic Contribution

Currently, there is little research available discussing the attributes that drive loyalty program participation, and even less in focussed on the restaurant industry. By researching customers motivations to participate in a loyalty program, it will advance the academic understanding of this underexposed aspect of the widely embraced notion of customer loyalty. In addition, the current study builds upon and integrates different sources of current knowledge on loyalty program participation. This will optimize existing understanding of the subject and enables the author to integrate it into a more advanced theoretical framework. Lastly, the majority of studies discussing loyalty program participation are conducted in the retail industry. The current study will apply an extended existing theoretical framework to the restaurant industry. By testing the effects in a new domain, the author contributes to the validation of the existing framework and its generalizability across industries.

1.3 Managerial Implications

(11)

11

industry, and loyalty programs are proven to foster customer loyalty. Given the large absence of loyalty programs in the restaurant industry, the author provides restaurant managers with a logical first step in creating a restaurant loyalty program: What attributes are suitable and valued by customers in a loyalty program?

(12)

12

2. THEORETICAL FRAMEWORK

This chapter defines the theoretical constructs and places it into context of the current research. Since customer loyalty encompasses the field of research of the current study, the definition of customer loyalty will be given, followed by its differences in the restaurant industry. The next section will elaborate on the characteristics of customer loyalty programs Thereafter, the framework of De Wulf et al (2003) will be used as a starting point for identifying attributes that drive program participation. Since the framework is based on the notion of equity theory, this concept will be discussed on forehand. Subsequently, the framework will be adapted to a restaurant setting and extended with additional variables found in literature.

2.1 Customer Loyalty

Customer loyalty has been a topic that has been extensively studied over the past three decades, and has given rise to a wide variety of definitions. Currently, the general contention is that customer loyalty is a behavioural measure which can be characterised into two types: (1) Attitudinal loyalty and (2) Behavioural loyalty (Kumar & Shah, 2004; Uncles, Dowling & Hammond, 2003; Meyer-Waarden, 2008; Dick & Basu, 1994). Attitudinal loyalty can be characterized as a positive perception and emotional attachment one has with a brand or store which promotes a repeatedly favourable response towards a product or brand of a firm. (Jahanshani, Hajizadeh, Mirdhamadi, Nawaser & Khaksar (2014). Behavioural loyalty can be described as loyalty from a customer as observed by one’s actual purchase behaviour, with typical metrics such as share-of-wallet or and share-of-purchase (Kumar & Shah, 2004; Jahanshani et al., 2014)

(13)

13

not building true, sustainable loyalty. To elaborate on the previously given example of supermarket consumer X: If a new supermarket would open which offered even less traveling time than the current supermarket, the absence of relational exit barriers formed by attitudinal loyalty (Shapiro & Vivian, 2000) is likely to cause consumer X to churn towards the more conveniently located supermarket.

Therefore, for a firm opting to create sustainable and profitable “true” loyalty, one must focus on establishing and maintaining a combination of preferential behavioural and attitudinal loyalty among its customers (Engel & Blackwell, 1982; Kumar & Shah, 2004). Without this concurrent approach, customer loyalty is only short-term oriented and does not utilize its full potential (Demoulin & Ziddah, 2009; Jahanshani, 2014). Yet in contrast, Kumar & Shah (2004) state that when a firm actively manages “true” loyalty and in conjunction with profitability, it could function as one of the most potent weapons in the marketing arsenal.

2.2.1 Customer Loyalty in the Restaurant Industry.

While consumption behaviour can be based on functional, utilitarian motivations (e.g. one needs to eat and drink in order to survive), fine dining restaurants try to satisfy hedonic or emotional motivations of customers by focussing on the total customer experience (Arnold & Reynolds, 2003; Lin, 2004; Ryu & Jang, 2007). In line with Verhoef et al., (2009), the customer experience involves the cognitive, affective, emotional, social and physical responses to the firm. This means that people do not only go to restaurants to quench thirst or still hunger, but seek to satisfy other needs such as interacting with friends just creating a memorable experience away from home (Ryu & Han, 2011). Subsequently, the design of the customer experience in the restaurant industry is often a key antecedent of both behavioural and affective restaurant loyalty (Matilla, 2001; Bowden & Dagger, 2011) Many studies has found that generating a high-quality customer experience is one of the central concepts in how restaurants create loyal customers and create a sustainable competitive advantage (Berry, Carbone & Haeckel, 2002; Berry, Wall & Carbone, 2006; Verhoef et al., 2009).

In an answer to the positive effects of a well-designed customer experience on customer loyalty, substantive research has revealed the key components which drive customer experience in the restaurant industry (Palmer, 2010; Jin, Lee & Huffman, 2012; Ryu, Lee & Kim, 2010; Harrington, Ottenbacher, Staggs & Powell, 2012; Ryu & Han, 2010) which are atmosphere, food quality, service quality and price fairness.

(14)

14

Atmosphere

The atmosphere, also often referred to as environment or service scape, includes many aspects. Among many, examples are cleanliness, scent, lighting, music, colours and furniture. Bitner (1992) defined the former as the servicescape, which is the built man-made environment as opposed to the natural or social environment. The servicescape can be segregated into three components: (1) Ambient conditions (scent, temperature, volume), (2) Spatial lay out/functionality (table setting, location) and (3) Signs, symbols & artefacts (restaurant logo). Another aspect which influences the atmosphere is the level of crowdedness in a restaurant (Ha, Park & Park, 2016) In a crowded restaurant, people can get agitated by the volume of the chatter. However, a busy restaurant also signals a good reputation whereas a quiet restaurant signals the opposite (Tse, Sin & Yim, 2002).

Atmospherics have the important potential of creating a positive impression that directly enhances customer experience (Berry et al., 2006). In addition, strong evidence exists that atmosphere has a direct link with customer loyalty (Kim & Kim 2004, Liu & Hang, 2009, Harrington, Ottenbacher, Staggs & Powell, 2012) However, it also imposes a danger. Customers are more likely to remember the negative parts of the atmosphere as opposed to the positive positive aspects of the dining experience (Chung & Hoffman, 1998; Clark & Wood, 1998; Matilla, 2001; Sulek & Hensey, 2004)

Food Quality

Food quality is the most important factor of the restaurant experience that drives restaurant patronization, and has a direct link with customer loyalty (Namkung & Jang, 2007; Sulek & Hensley, 2004; Clark & Wood, 1996). It is essentially the core product that a restaurant provides, and is evaluated by customers among many dimensions. Examples are nutritional value, appeal, tastiness, variety, freshness and temperature. (Kivela et al., 2000; Jin, Lee & Huffman, 2012).

Service quality

Literature has proven that service quality is one of the key determinants that shape the customer experience in a restaurant setting (Matilla, 2001; Kattara, Weheba & El-Said 2008; Stevens, Knutson & Patton, 1995).

(15)

15

responsiveness, competence, access, courtesy, communication, credibility, security, understanding the customer and tangible aspects such as appearance of the personnel. Service quality greatly relies on employee behaviour – actions carried out by the organization that has an effect on the customer (Harrington et al., 2012). Service quality is directly affecting the customer experience, which influences customer satisfaction (Bolton & Drew, 1991; Gustafsson, Johnson & Roos, 2005, Ryu, Lee & Kim, 2010)

Price Fairness

(16)

16

2.3 Customer Loyalty Programs

An increasingly important strategical component of many companies to foster customer loyalty is the customer loyalty program (Leenheer et al., 2007; Demoulin & Zidda, 2008; Nunes & Dreze, 2006; Uncles, Dowling & Hammond, 2003).

This study adopts the definition of loyalty programmes as given by Dororic, Bijmolt & Verhoef (2012) and combines it with Shoemaker & Lewis (1999): A loyalty program is a formal program of customer relationship management with the following dimensions.

Fosters loyalty:

The main purpose of the loyalty programme should be to foster both attitudinal and behavioural loyalty, resulting in both affective and transactional outcomes for a firm.

Structured:

A customer must formally become a program member to obtain the benefits. This provides the customer with benefits in terms of discounts but enables the firm to “get to know” the customer based on information given.

Long-term:

The program has a long-term scope, as it focusses on long-term benefits for both the firm and the customer.

Rewarding:

A loyalty programme offers the participant rewards such as discounts or non-monetary rewards (e.g. preferential treatment). The rationale behind offering rewards to the customers is that the firm is able to receive additional value brought in by the loyalty of the customer (Kumar & Shah, 2004)

Ongoing marketing efforts:

The firm should adapt its marketing efforts to fit the demands of the members (e.g. targeted vs non-targeted marketing offers) on an ongoing basis.

(17)

17

perceived as additional value from a customer perspective, it will increase the probability that a customer evaluates the firm positively (Dowling & Uncles, 1997).

Currently, a threat exists regarding the general scope of the majority of loyalty programs, which focus on behavioural loyalty (Kumar & Shah, 2004). An example is the ubiquitous stamp card at many coffee bars which offer a free cup of coffee after 10 purchases - “buy 10, get the 11th for free". As this might stimulate repeat patronage, it is transaction-based which does not necessarily invoke attitudinal loyalty (Yang & Peterson, 2004). Among other reasons, this has led to a debate in current research. Past research has praised the potential value of loyalty programmes (Nunes & Drèze, 2006; Meyer-Waarden, 2007), while others have questioned it (Downing & Uncles, 1997; Sharp 2010). As a critique, many programs have not been successful because of a lack of focus on how members perceive loyalty programs and which advantages individuals derive from participation (Mimouni-Chaabane & Volle, 2010).

(18)

2.4 Conceptual Model

Participation Exclusivity

Participation Efforts

Program Benefits

Number of Program Providers

Program Duration Personal Data Investment

(19)

19

2.5 Loyalty Program Participation

Past research (Demoulin & Zidda, 2009; De Wulf et al., 2003; Leenheer et al., 2007) explains that a customer’s decision to participate in a loyalty program is based on the concept of equity theory. In case of inequity, the benefits/costs in an exchange relationship are inconsistent with the perceived benefits/costs of the outcomes of the exchange relationship (Leventhal, 1980) In other words, inequity can result in dissatisfaction of both exchange partners, which are subsequently expected to adapt future exchange behaviour to prevent additional dissatisfaction by means of smaller investment in the relationship (Leventhal, 1980). On the contrary, when equity prevails, both exchange partners invest a constant amount of inputs in the relationship, which results in the satisfaction of exchange partners with their outcomes (Leventhal, 1980; Lewin & Johnston, 1997).

Placed into context, the decision to participate in a loyalty program depends on the balance between the investment that a customer must offer in return for the investment from the firm. Both parties will have to give up something, of which the costs are expected to be outweighed by the advantages (De Wulf et al., 2003). For example, the participant gives up some form of personal data in exchange for a discount on its subsequent purchase. The program provider will give up some potential margin to invest in a long-term relationship. Thus, the likelihood of participating in a loyalty program depends on how much effort a customer must put in to obtain the rewards the program provides (Dorotic et al., 2012; Kivetz & Simonson, 2003; Lee, Kim & Lee, 2013).

2.6 Loyalty Program Attributes

Following De Wulf et al., 2003, the balance between the effort and results one expects from a loyalty program, and thus determining participation is based on eight attributes. These attributes can subsequently be classified into two categories

2.6.1 Customer inputs

Personal data required  Defined by De Wulf et al., 2003 as “personal data consumers are

(20)

20

Gerstmeier, Tafreschi & Enzmann & Schneider, 2007) However, in the light of equity theory, one might still participate if the costs of giving up personal data are outweighed by the benefits of the program (van Doorn, Verhoef & Bijmolt, 2007). Therefore, the author hypothesises:

H1: The likelihood of participating in a loyalty program increases when the amount personal data required decreases.

Yet, recent privacy scandals and mismanagement of information have led to more privacy concerns among customers. Data collection by itself often triggers privacy concerns of customers (Bansal & Zahedi, 2008; Malhotra, Kim & Agarwal, 2004). A customer’s privacy concerns are defined as an individual’s subjective views of fairness about how a company treats personal information, and can range from high to low (Bansal & Zahedi, 2008; Campbell, 1997). This implies that not all customers are equally concerned about the way companies treat and use the data one provides a company with (Phelps, Nowak & Ferrell, 2000). Some customers carefully examine the application of data within a loyalty program before giving up personal information, while others carelessly disclose personal information as they see it as an unavoidable prerequisite (Awad & Krishnan, 2006; Hinz et al., 2007; Jai & King, 2016). Furthermore, Hinz et al., (2007) state that privacy concerns are especially important within loyalty programs. As nearly all loyalty programs require some degree of data disclosure, one can reason that customer’s privacy concerns are likely to affect the amount of information one accepts to disclose when joining a loyalty program. Thus, the author hypothesises:

H1a: The negative effect of personal data required on the likelihood of participating in a loyalty program is stronger when the customer privacy concerns increase.

Participation exclusivity  The extent to which participating in the loyalty program is

(21)

21

being higher-status. Thus, in accordance with Arbore & Estes, 2013 and De Wulf et al., 2003, the author hypothesises:

H2: The likelihood of participating in a loyalty program increases when the program has certain entry requirements.

Participation efforts  The activities a loyalty program member must undertake to obtain the

benefits of the program (De Wulf et al., 2003). In case the loyalty program requires the customer to put a lot of time in administration or repeatedly show one’s member-card, the satisfaction with the obtained benefits decreases (Cardozo, 1965). In the same regard, a lot of customers who want to enrol in a loyalty program base one’s decision on the balance between the costs and the benefits it offers (Demoulin & Zidda, 2009; De Wulf et al., 2003; Leenheer et al., 2007) Furthermore, if a customer must invest more in the loyalty program than the firm, the notion of equity theory explains customers will more likely to be dissatisfied with the results (Leventhal, 1980). Therefore, the author hypothesises:

H3: The likelihood of participating in a loyalty program decreases when the amount of participation effort required increases.

2.6.2 Reward Types

Program benefits  The rewards that are returned to the program member in exchange for the

(22)

22

H4: Utilitarian benefits have a stronger influence on likelihood of participating in a loyalty program compared to hedonic and symbolic benefits.

Number of program providers  Defined by De Wulf et al., 2003 as the number of vendors

supporting a single loyalty program. Potential program participants may be attracted to the fact that a single program can be used for multiple vendors, and thereby offer greater perceived value compared to a standalone program (Cappizi & Ferguson, 2005; Dorotic, Fok, Verhoef & Bijmolt, 2010). An example of multi-vendor programs is a loyalty program that can be used at multiple firms (e.g. Airmiles). Because a loyalty program for multiple vendors allows participants to collect points faster, increased convenience and more redemption options, it is considered to offer greater value than a single program. (Cappizi & Ferguson, 2005; Dorotic et al., 2010). Based on the former the author hypothesises:

H5: The likelihood of participating in a loyalty program increases if the program can be applied in multiple restaurants.

Program duration  Defined by De Wulf et al., 2003 as the period of time in which the benefits

of the program are available to the program participant. Dorotic et al., (2012) found that in many industries a loyalty program which is of short duration and has a fixed end-date are most popular among firms. However, it remains unknown if customers prefer a short-term program compared to a long-term program. Since accumulating points in a loyalty program has an important psychological benefit to customers (Liu, 2009), one could say that this implies that sustaining a long-term program leads to more opportunities for customers to experience this psychological benefit. On the other hand, short-term programs often include reward expiration which may invoke customer frustration, or so-called point pressure (Dorotic et al., 2014). To avoid customer frustration from point pressure, many companies also apply long-term loyalty programs, in which there is no expiration date of the rewards. Furthermore, O’Brien & Jones, (1995) and Lewis (2004) explain that a major factor that customers consider when evaluating a loyalty program is the likelihood of receiving a reward and time constraints. When a program is long-term, a customer has more time to save points or redeem rewards as opposed to a program with a short-term duration. Therefore, the author hypothesises:

(23)

23

2.7 Influence of Variety seeking Orientation

It is known that loyalty programs focus on repeat patronage and fostering loyal customers. However, research have shown that the variety seeking orientation of customers plays a strong role in restaurant choice (Yung & Hoon, 2012; Kahn, 1995). Variety seeking orientation in this context is defined as the instance of dining at a restaurant reduces the probability that one would visit the same restaurant again. Furthermore, Jung & Hoon (2011) state that the variety seeking orientation of a restaurant customer is important in determining whether to visit a new or previously visited restaurant. If a customer has a low variety seeking orientation, this means that if one is satisfied with a restaurant one is more likely to return. In a case of high variety seeking orientation, the customer chooses for another restaurant – not because one is not satisfied with the focal restaurant - but because one wants to experience something new - or the hedonic motive is stronger compared to the utilitarian (Holbrook & Hirschman, 1982; Hoyer & Ridgway, 1984). Since the restaurant industry focusses on the experience and thus more on hedonic motives (Arnold & Reynolds, 2003; Lin, 2004; Ryu & Jang, 2007), it is likely that the variety seeking orientation moderates the influence of the loyalty program attributes on program participation. If one is not planning to return to the same restaurant because one seeks diversity, one is unable to receive benefits from a loyalty program for repeat patronage and might therefore choose not to participate. An example is a point-based reward program. If one cannot save up rewards over time because one likes to visit different restaurants, such a program would have no value. However, if a program offers pre-planned hedonic benefits such as a workshop or food tasting event which is open to all loyalty program members, one can readily benefit because there is no consecutive effort required.

Furthermore, Zang, Krishna & Dhar (2000) found evidence that the type and timing of rewards offered by the loyalty program are important for high-variety seeking customers. Therefore, the author believes that a variety seeking customer is not able to fully obtain benefits from a loyalty programme, which influences the decision to participate. Thus, the following is hypothesized:

H4a: The influence of the type of reward on loyalty programme participation decreases when a customer has a high variety seeking orientation

(24)

24

obtaining the program rewards is less limited by visiting different restaurants. Based on the former, the author hypothesises:

H5a: The effect of a multi-vendor loyalty program on loyalty programme participation increases when a customer has a high variety seeking orientation.

2.8 Control Variables

Participation fee  Defined by De Wulf et al., (2003) as “a certain amount of money that has

to be paid in order to be able to participate in the loyalty program”. Next to the amount of personal information that must be disclosed, some loyalty programs require a monetary fee in order to participate. Yet, customers seek to minimize costs in order to maximize the benefits of the program (Lee, Kim & Lee, 2013). Furthermore, De Wulf et al., (2003) found evidence for this claim by stating that people prefer to pay no participation fee as they consider the program offering as an appreciation for their loyalty and not something one should pay for. In line with this theory, many loyalty programs nowadays do not require a participation fee but are free of charge. Therefore, Participation fee is excluded from the participation drivers but will be controlled for in the study design.

Purchase frequency  Defined as the number of purchases per customer in a specific

time-period (Sharp & Sharp, 1997). It has been found that an increased purchase frequency decreased the perceived convenience costs, which in turn influenced the customers decision to enrol in a loyalty program (Meyer-Waarden 2007; Allaway, Berkowitz & D’Souza, 2003) Furthermore, De Wulf et al., (2003) state that when a customer visits a firm more often one can generally receive more benefits from a loyalty program. This would imply that people who dine out more often are likely to receive more value loyalty program. In addition, Galguera, Luna & Mendez (2006) also found an effect of shopping frequency influencing loyalty program participation. Therefore, purchase frequency is controlled for in the study design.

Age  Another customer characteristic of which is it not yet known if it influences loyalty

(25)

25

3. METHODOLOGY

The research methods and related specifics are discussed in the current chapter. First, the research method is explained together with the according scale. Secondly, the moderator measurements are discussed. Conclusively, control variable implications on the research design are explained in the last part.

3.1 Restaurant in the Current Study

The author makes a distinction in this study between restaurants and fast-food restaurants. People often visit fast food restaurants for convenience and low costs (Glanz, Basil, Maibach, Goldberg & Snyder, 1998). When the report refers to restaurants, it refers to a casual or fine-dining restaurants in the Dutch market which try to satisfy hedonic or emotional needs by focussing on a positive experience (Arnold & Reynolds, 2003; Lin, 2004; Ryu & Jang, 2007).

For the survey, the author sketches a restaurant which realistically represents an average restaurant in the Dutch market. This restaurant has an average three-course menu price of € 27,45 (Misset Horeca, 2014) and has 3 locations throughout The Netherlands. In addition, the restaurant sketch purposely does not inform the respondent about the type of cuisine, as this might bias the answers.

3.2 Method

(26)

26

include a no-choice option which respondents can choose if they would not accept any of the alternatives, which increases realism (Eggers & Sattler, 2011).

3.2.1 Attributes and Levels

The most important factor in a successful conjoint analysis entails creating accurate and realistic attributes with corresponding levels. In this case, the attributes and levels should realistically portray how a loyalty program could look in the restaurant industry. To reduce complexity and fatigue for respondents, the maximum number of levels for a choice-based conjoint analysis is six (Eggers & Sattler, 2011). This allows for testing all attributes which can be found in the conceptual model.

Furthermore, on average the number of levels is about three to five levels per attribute. In addition, more levels increase the number of parameters one must estimate which subsequently requires more data (Eggers, 2016). Given the limited time frame of the current research, three levels are used per attribute. Based on the literature and theoretical overview of this report, the following set of attributes and levels has been created. Note that the attributes are phrased somewhat different compared to the variables in the theoretical overview to avoid complexity in the choice design.

Table 1. Attributes and levels

Level 1 Level 2 Level 3 Personal data investment Program activated by providing Anonymous registration code E-mail address & Age Name & address Participation exclusivity Program entry requirements No entry requirements Visit the restaurant 3 times Personal invitation by restaurant Participation efforts Participation efforts

(27)

27 Number of program providers Program accepted at My nearest "Am" location All locations of "Am" All locations of "Am" + 3 other restaurants Program duration Program duration Lasts for 6 months

Lasts for 1 year Has no end-time.

Program activated by providing (Personal data investment)

the first level is “Anonymous registration code”. In this case, no data is required but the card is activated by registering it with a code given by the restaurant. For example, the restaurant gives you a card which has a 5-digit code printed on the back. The restaurant also gives you an activation code for one’s card, which can be filled in on the restaurant’s website to activate the card. In this case the card is anonymous and the level of data invested is low. Level one and two of this attribute require increased personal information and thus represent increased levels of personal data investment.

Program entry requirements (Participation exclusivity)

The first level represents a program which has no restrictions and is open to anyone. The second level places some additional restrictions on participation, but these are still achievable for most customers. Level three represents a loyalty program of which the members are hand-picked. This is likely to create a hierarchy among customers (Arbore & Estes, 2013) in which the members of the loyalty program feel more “special” than regular customers, contributing to one’s propensity to participate in the program.

Participation efforts (Participation efforts)

 At level one, a program member only has to scan one’s card at every visit to register the check-out/save rewards et cetera. At the subsequent two levels the program member must fulfil an additional requirement – periodic validation of one’s loyalty card – which requires additional efforts.

Program rewards (Program benefits)

(28)

28

Program accepted at (Number of program providers)

Level one represents a program that can only be utilized by customers of restaurant X located

in your nearest city. Level two increase the number of providers to all restaurant locations of restaurant X. Finally, level three expands the application of the program to a multiple-restaurant program. This means that the program can be utilized at different restaurants in different cities.

Program duration (Program duration)

The first level represents a fixed end time of a loyalty program on a relatively short notice. The second level also has a fixed end time but extends the duration of the program. The third level represents a program without a fixed end time.

3.3 Designing the Choice Sets

The design of the choice sets in the conjoint analysis should adhere to a fundamental requirement to increase the quality of the results: Provide a limited amount of choice sets to avoid respondent fatigue (Eggers & Sattler, 2011). To determine the optimal number of choice sets, the author follows the formula provided by Hair, Black, Babin &Anderson (2010). Minimum number of choice sets = total number of levels across all attributes – number of attributes + 1.

This concludes into 13 choice sets, which is a suitable amount (Eggers & Sattler, 2009; Eggers & Sattler, 2011). The attributes and levels within the choice sets will be randomized to prevent bias of the results based on ordering.

Since there’s no literature consensus about what the optimal number of alternatives per choice set should be, the author follows (Eggers, 2016), stating that the number of alternatives should be between two and five. The author will therefore provide the survey respondents with three alternatives per set. In addition, a no-choice option has been included to increase realism (Eggers & Sattler, 2011).

3.4 Measurement for Moderating Variables

3.4.1 Privacy Concerns

(29)

29

of this rather small scale is that would have less influence on the fatigue and motivation of a respondent to continue, compared to a scale with more items (Eggers & Sattler, 2011; Robinson, Shaver & Wrightsman, 1991). This is beneficial for the reliability and validity of the choice-based conjoint analysis, which is conducted after measurement of the moderating variables. The scale is measured on a 5-point Likert scale which results in a privacy concerns measurement ranging from (1) Weak privacy concerns to (5) strong privacy concerns. The scale can be found in the upcoming table.

Table 2. Privacy concerns scale

Item Measurement

Loyalty cards form a threat to my privacy due to their registration system. 5-point Likert I do not like companies to obtain information on my purchase behaviour. 5-point Likert

3.3.2 Variety seeking Orientation

The variety-seeking orientation of restaurant customers will be tested using the scale implemented by Jung & Yoon (2012). This scale has been successfully applied in a restaurant loyalty setting and is therefore viewed as suitable for the current research. The six-item scale is measured on a 5-point Likert scale and results in a measurement ranging from (1) weak variety-seeking orientation to (5) strong variety seeking orientation. The scale can be found in the table below.

Table 3. Variety-seeking orientation scale

Item Measurement

I like trying new things more than doing familiar things 5-point Likert

I want to try new restaurants 5-point Likert

I enjoy taking chances with unfamiliar restaurants just to get some variety 5-point Likert When I go to a restaurant, I feel it is safer to order dishes I am familiar with 5-point Likert I would rather stick with a restaurant I usually visit than try something I am not

very sure of

5-point Likert

I am very cautious when trying new or different products 5-point Likert

3.5 Control Variables

(30)

30

Purchase frequency is measured by a multiple-choice question with 5 levels. Based on khn (2012), Dutch people dine out approximately twice a month. Therefore, the author measures the purchase frequency of respondents in a period of one month. The question is “per month, I usually visit a restaurant x times.” and has the following answer possibilities.

(a) Never (c) 2 times (e) 4 times

(b) 1 time (d) 3 times (f) 5 times or more

Furthermore, in line with equity theory (Leventhal, 1980), a free loyalty program would be preferred over a loyalty program that requires an entry fee. The author assumes that this is a reason why most of nowadays loyalty programs are free of charge to participate in. To control for the effect of an entry fee, the introduction text of the conjoint analysis portrays a loyalty program and explicitly mentions that this program does not require an entry fee. This is in line with Eggers & Sattler (2011): If attributes are relevant to the research but are not the primary focus, they can be included but remain fixed.

3.6 Data Collection

In order to gather data to test the hypotheses, a survey has been created on www.preferencelab.com. The link to the survey will be included in the appendix (on page x), and a sample of the questions can be found on appendix page x. The survey will be distributed via social media, internet forums and to peers of the author. The survey starts with an introductory text, after which the respondent is asked to fill in personal details. Thereafter, the choice-based conjoint survey is presented, followed by questions measuring the moderating effects.

(31)

31 Both surveys can be found in Appendix B.

A small pre-test (n=10) with respondents from Norway and The Netherlands identified some survey flaws, which have been corrected before launching the definitive version.

3.7 Plan of Analysis

A total of 153 respondents have filled in the survey questions over a period of two weeks in order to prevent data dilution, only the respondent entries which have completed the full survey were kept (N=147).

3.7.1 Model Specification

Conjoint analysis is based on modelling the utility of a specific alternative (U), in this case (Ui)=loyalty program. The implicit assumption of this method is that consumers choose

alternatives that yield the highest utility (Eggers, 2016).

For modelling the conjoint survey data, the software package Latent Gold was used. To measure the effect of each attribute, separate parameters have been estimated. Secondly, the measurement coding of the attribute levels has implications for the results of the choice-based conjoint. In this study, effect coding was used. Effect coding means that the reference level is set to -1, which results in more realistic effects of attribute levels (Moore, 1980). This is different from regular dummy coding, where the reference level is set to 0. In the case of this study, where attributes have 3 levels, it means that the reference attribute level is coded -1, the second one is coded as 0 and the third is coded as 1.

The model estimation technique is multinomial logistic regression. Utility in itself is not a proper estimate of choice (Eggers, 2016). Instead, one needs to transform the utility function (U) in a probability model. Doing so allows the author to infer the probability of choosing a certain loyalty programme between 0-100%.

Furthermore, an aggregate choice model will first be estimated. Thereafter, to account for customer heterogeneity in modelling, a latent class analysis will be conducted to estimate models accounting for differences in restaurant customer preferences.

3.7.2 Model fit and Validation

(32)

32

AIC and BIC. AIC and BIC help determining the optimal model by penalizing for “unnecessary’ model parameters.

To validate the model, the hit rate is used as a measure of accuracy as prescribed by Eggers (2016).

3.7.3 Moderator (control) Validation and Analysis

For estimation purposes, the scales used to measure the moderating effects will be summarized into factors. However, to do so, the author first needs to test whether the constructs in the survey accurately measured the moderating concepts and can subsequently be combined into one dimension. Therefore, the author makes use of factor and reliability analysis.

Cronbach’s alpha will be utilized to test the reliability of the factors. Factors with an alpha >0.6 will be regarded as adequate. Furthermore, following Williams, Onsman & Brown (2010), Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy and Bartlett's Test of Sphericity will be applied to test if the sample is adequate. If the scales prove not to be reliable, the author will rely on dimension reduction by means of factor analysis. Herein, the number of factors will be determined by comparing the eigenvalues of the separate factors, and deducting the items that do not meet the threshold value of >1.

(33)

33

4. RESULTS

The first step is to give an image to the sample set. N=153. After validation, 6 respondents were filtered out as they answered the survey with systematically choosing the none-option which potentially biases the results. This resulted in a final sample of N=147.

Table 4. Sample description overview

Age

_{4.1 Sample Description}

The survey was conducted among the peers of the authors and subsequent “peers-of peers”, which are many students. Therefore, the respondents showed a diverse age, but the mean age is as expected relatively low. Furthermore, almost an equal number of males and females participated in this study, of which the large majority is Dutch.

In addition, the majority of respondents are “average spenders” when they visit restaurants (between €10 and €40) with very little people spending less or more (5%). When asked what determines one’s restaurant choice, the sample shows that atmosphere, food quality, price and location respectively are most important. Very little observations of the “other” option were measured. Additionally, table 13 in appendix A shows that most respondents (66%) prefer a restaurant with a French cuisines over other types of cuisines such as Asian cuisine (9%).

Concluding, the largest group of respondents (44%) visits a restaurant 1 time per month. The second largest group (26%) visits 2 times per month, whereas the smallest group of restaurant visitors (3%) never visits a restaurant – at least not on a monthly basis.

Mean Range 29 years 15-65 years Gender Percentage male Percentage female 52% 48% Nationality Amount of Dutch Other European Other non-European 92% 7% 1%

What is the average spending per person at each visit Less than €10 €10/€20 €20/€30 €30/€40 €40/€50 More than €50 1% 20% 34% 30% 11% 4%

(34)

34

4.2 Reliability Analysis of Moderating Variables

To test whether the scales of the moderating variables are reliable, the author validated the items via factor analysis and Cronbach’s alpha. If the scales are reliable according to Cronbach’s alpha (alpha >0.6), the author can assume that the existing scales are considered validated and applicable for further use. In addition, Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy and Bartlett's Test of Sphericity will be applied to test if the sample is (1) likely to factor well (KMO >.5) and (2) variables are related (should be significant) First, the scale measuring privacy concerns was evaluated and is summarized in the table below.

Table 4. Reliability test of moderating variables

(Sub)Item Cronbach’s alpha Alpha if deleted N of items

KMO Bartlett Mean value (0-4) Privacy concern .694* 2 .500* P <0.01* 1.98 PC1 - 1.89 PC2 - 2.07 Variety seeking orientation .766* 6 .720* P<0.01* 2.52 VSO1 .713 2.40 VSO2 .749 3.23 VSO3 .740 2.77 VSO4 .713 2.01 VSO5 .734 2.21 VSO6 .737 2.51

(35)

35

removing an item would not improve the scale. Thus, a new factor has been distilled which is also a standardized variable.

4.3 The Aggregate Choice Model

Now that the sample is known and the dataset has been prepared with validated and standardized moderating variables, the choice model can be estimated using Latent Gold. The first section starts with estimating an aggregate choice model. However, an aggregate model does not fully account for differences between respondents. Thus, to account for respondent heterogeneity, a latent class analysis will be performed in the latter section in which segments and coherent models have been estimated.

The first step in choice modelling is to test for potential linearity within the attributes (Eggers, 2016). However, the design of the attribute levels in the current study do not allow for linearity. The attribute levels are all nominal (categorical) and therefore linearization is not possible (Eggers, 2016). Thus, the attributes are considered as part-worth coefficients in the modelling process.

4.3.1 Model Comparison

In order to determine the optimal model, a model comparison has been performed which evaluates multiple tests and concludes with the optimal model.

Table 5. Model comparison measures

Model number (Progressive)

(36)

36

Table 5 on the previous page gives an overview of the comparison criteria. Model 1 is a basic model with only the 6 attributes of the conjoint analysis included as variables. Model 2 is a model with the 6 attributes and all moderating variables. However, model 2 yielded a lot of insignificant estimates and it showed a lower fit compared to model 1. Still, the author is interested in the moderating effects and therefore, multiple models are estimated based on the notion of backwards elimination with model 2 as a starting point (Yuan & Lin, 2006). This entails that the least significant estimate (highest p-value) was taken out and a new model was estimated excluding this estimate until all estimates were significant. Note that all main effects were included in every model because removing a main effect would yield an unrealistic model (choices are always made as a trade-off between attributes (main effects), therefore the insignificant attributes still contain information).

When comparing information criteria, one can see a decreasing BIC, AIC and AIC3 when progressing (removing the least significant variable) from model 2 through 7. Thus, in terms of information criteria, model 7 fits the data best.

Secondly, the likelihood ratio test between all models and the null models prove to be significant, which entails that all models perform better than the null model. Subsequently, a likelihood ratio tests have been performed between models in a progressive order (comparing the model with the least significant estimate removed with the previous model where the estimate was still included). By doing so, the author can validate that making the model more parsimonious does not drastically change the model fit. As one can see, the p-values are highly insignificant, which means that removing the insignificant variable does make the model more parsimonious while maintaining the same amount of estimation power. The hit rate and McFadden R2 _{remain constant with an acceptable hit rate (63%) and a rather low McFadden R}2

(37)

37

Ultimately, model 7 was estimated (table 13 in appendix A) and proven to be the best model. Model 7 includes all six attributes included and one moderating variable – the effect of variety seeking orientation on the number of program providers.

4.4 Model Interpretation

When interpreting the model, the main effects are most important and therefore discussed first. Following the main effects description is the interpretation of the moderating effects. While discussing the main and moderating effects, hypotheses have been tested and results can be found in the text as well as in table 13 of appendix A.

4.4.1 Main Effects

As explained before, model 7 is the final, best fitting model and is used to explain the main effects. The main effects are displayed in table 6 which entails the attributes and corresponding levels, utility estimates, Wald statistics, corresponding p-values and the relative importance.

Table 6. Model 7 main effects

Attributes Utility Wald p-value Relative importance Personal data investment 12.13 p<0.01

Anonymous registration code 0.1039

3% E-mail address & Age -0.0775

Name & Address -0.0264

Participation exclusivity 63.40 p<0.01

No requirements 0.2277

18% Visit the restaurant 3 times -0.2024

Personal invitation by restaurant -0.0253

Participation efforts 81.21 p<0.01 Scan loyalty card at payment 0.2494

20% Scan loyalty card at payment + fill in 1

minute survey

-0.0046 Scan loyalty card at payment + 3

minute survey

-0.2448

Program benefits 717.46 p<0.01 Discount on food/drinks 0.7666

50% Special food workshops -0.3202

Priority when reserving a table -0.4464

Number of program providers 1.43 p=0.49 Nearest location of restaurant X 0.0537

(38)

38 All locations of restaurant X + 3

similar restaurants

-0.0208

Program duration 7.37 p=0.03

Lasts for 6 months -0.0761

6%

Lasts for 12 months 0.0044

Has no end time 0.0717

None option 55.31 p<0.001 -0.4576

The Wald statistic explains that the attribute estimates are all significant at the level p<0.05 or p<0.01, except for number of program providers. Program benefits is the most important attribute in the used sample by making up 50% of the importance when choosing.

Participation efforts (20%) and participation exclusivity (18%) are two attributes of

significant yet smaller importance important. The attributes of low importance are program

duration (6%) and personal data investment (3%). Respectively, number of program providers

is insignificant and its effect and corresponding importance cannot be interpreted. The estimate for the none option (no loyalty program is preferred in general) with β=-0.4576 and p<0.01 shows that in general people prefer to join a loyalty program consisting of a

combination of the attributes used in this study instead of not joining a program at all). Continuing, the attribute estimates will be discussed. As mentioned, the p-values of the attributes mean that the attributes are significant and different from zero. However, no information is available which explains if the attribute level estimates are statistically different. Within a balanced choice design, it is possible to evaluate the statistical difference by calculating a t-statistic by dividing the mean difference by the standard error of the differences (Hauber et al, 2016). An alpha of 5% was maintained in this study to evaluate the t-statistic. First, H1 assumes that the likelihood of participating in a loyalty program increases when the amount personal data required decreases. To investigate this hypothesis, the attribute personal

data investment must be significant which is confirmed (p<0.01). Since anonymous registration

(β= 0.1039) code is the lowest level of personal data required, the other two levels should offer lower utility. The utility of anonymous registration code is significantly higher than the utility of having to give up e-mail address & age (β= -0.0775; t=2.89, p<0.05) or name & address (β= -0.0264; t=2.09, p<0.05). Therefore, one can conclude that a lower amount of personal data required by a loyalty program is preferred and H1 is thereby confirmed.

Secondly, H2 assumes that the likelihood of participating in a loyalty program increases when

(39)

39

to be significant (p<0.01). The attribute level personal invitation by restaurant (β=-0.0253) is considered to elicit the highest level of entry requirements and is therefore taken as benchmark. However, the estimates reveal that no entry requirements offers the highest utility (β=0.2277). The t-test revealed that the difference in utility between no entry requirements and personal

invitation by restaurant (β= -0.0253; t=4.07, p<0.05) and visit the restaurant 3 times

(β=-0.2024; t=2.76, p<0.05) is significant. Therefore, H2 is not supported but is confirmed in opposite direction.

Third, H3 expects that the likelihood of participating in a loyalty program decreases when the

amount of participation effort required increases. This holds if the attribute level scan loyalty

card at payment (β= 0.2494) has brings statistically higher utility. The general attribute is

significant (p<0.01). Furthermore, scan loyalty card at payment + fill in 1 minute survey (β=-0.0046;

t=4.10, p<0.05) and scan loyalty card at payment + 3 minute survey (β=-0.2448; t=7.78, p<0.05) are both offering lower utility and therefore H3 is confirmed.

Fourth, H4 assumed that utilitarian benefits have a stronger influence on likelihood of participating in a loyalty program compared to hedonic and symbolic benefits. The attribute is highly significant (p<0.01) and very important (50%). In addition, the utility of the utilitarian benefit – discount on food/drinks – is very high (β=0.7666) compared to the other hedonic

special food workshops (β=-0.3202; t=17.22, p<0.05) and symbolic priority when reserving a table (β=-0.4464; t=175.80, p<0.05). Thus, respondents prefer utilitarian benefits over other

benefits and H4 is confirmed.

Fifth, H5 predicts that the likelihood of participating in a loyalty program increases if the program can be applied in multiple restaurants. Since the attribute number of program providers is not significant (p=0.49), the hypothesis cannot be confirmed.

Sixth, H6 assumed that the likelihood of participating in a loyalty program increases if the program has no fixed end time. The attribute program duration is significant (p=0.03) and valid. The option no end time offers the highest utility (β=0.0717) and is statistically different from

lasts for 12 months (β=0.0044; t=2.35, p<0.05) but not from lasts for 6 months (β=-0.0716;

(40)

40

4.4.2 Moderating Effects

Following the interpretation of the main effects is the interpretation of the moderating effects. Only the effect of one’s variety seeking orientation has a significant interaction with the number of program

providers. The other moderating effects were insignificant in all the models tested presented and

described in table 5 (see table 18 in appendix A).

Moderating effect of variety seeking orientation on

Utility Wald p-value

All locations of restaurant X + 3 similar restaurants

0.1947 3.80 p=0.05

This points in the direction that respondents who have a relatively high variety seeking orientation tend to prefer the attribute level all locations of restaurant X + 3 similar restaurants, which is the highest level of number of program providers. However, the main effect of number of providers is not significant. Therefore, the exact interpretation of the interaction effect on its main effect is not reliable. Hence, H1a, H4a and H5a are not supported by the data.

4.4.3 Predictive Validity

To test the predictive validity of model 7, Eggers (2016) suggests evaluating the hit rate. The hit rate is a measure that evaluates how well a model can predict certain values. In this research, the hit rate of the final model (table 13 of appendix) is 63%, which is adequate.

4.5 The Segmented Choice Models

Now that an aggregate choice model has been estimated, much is known about the preference of the average consumer when joining a restaurant loyalty program. However, the aggregate model does not account for customer heterogeneity. Therefore, a latent class analysis is performed in the current section. A posteriori segmentation has been implemented to find out the optimal number of segments.

4.5.1 Determining the Number of Segments (classes)

(41)

41

Table 7. Latent class model comparison

model 8 (2 class) model 9 (3 class) model 10 (4 class) model 11 (5 class) model 12 (6 class) BIC 5914.655 5745.3609 5654.4624 5668.2295 5710.8377 AIC 5783.076 5524.0689 5343.4574 5267.5115 5220.4068 AIC3 5827.076 5598.0689 5447.4574 5401.5115 5384.4068 CAIC 5958.655 5819.3609 5758.4624 5802.2295 5874.8377 Classification error 0.0097 0.0156 0.009 0.0188 0.0179

When looking at BIC and CAIC it becomes clear that the model 10, the model with four classes, performs best. The classification error (0.009) also supports the former by showing the lowest error for a four-class model. However, AIC and AIC3 are optimal at a six-class solution and the graph (X) shows that these measures will keep decreasing (favouring a higher-class model). This contradiction is solved by applying literature. Andrews & Currim (2003) as well as Eggers (2016) describe that AIC and AIC3 generally favour more complex models. In addition, Andrews & Currim (2003) find that CAIC is the best measure for determining the number of segments. The elbow in the graph (x) as well as the classification error in table (x) support the contention that the most optimal number of classes is four. Therefore, model 10 is selected for further analysis.

4.5.2 Main Effect, Moderator and Covariate Interpretation

As mentioned, four segments have been found. An overview of the model results can be found in table 15 and 16 of appendix A. the segment and corresponding characteristics will be evaluated in the current section.

A first notion is that the attributes (main effects) personal data investment (p<0.01),

participation exclusivity (p<0.01), participation efforts (p<0.01), program benefits (p<0.01)

5000 5200 5400 5600 5800 6000 model 8 (2 class) model 9 (3 class) model 10 (4 class) model 11 (5 class) model 12 (6 class)

What drives Loyalty Program Participation? Choice-based Conjoint Analysis in the Restaurant Industry